Recommendation system for restaurants¶

Based on the Yelp Dataset.¶

0. Libraries¶

First of all, we define all the libraries we need.

from matplotlib.ticker import PercentFormatter as _PercentFormatter
import matplotlib.pyplot as _plt
import numpy as _np
import pandas as _pd
import joblib as _jl
import glob as _glob
import os as _os
import re as _re
import time as _time
from multiprocessing import Pool as _Pool
from sklearn.preprocessing import OrdinalEncoder as _OrdinalEncoder, binarize as _binarize
from sklearn.metrics import confusion_matrix as _confusion_matrix, roc_curve as _roc_curve, classification_report as _classification_report, accuracy_score as _accuracy_score
from sklearn.model_selection import GridSearchCV as _GridSearchCV
from sklearn.svm import LinearSVC as _LinearSVC
from sklearn.ensemble import RandomForestClassifier as _RandomForestClassifier
from keras.models import Sequential as _Sequential, load_model as _load_model
from keras.layers import Dense as _Dense

_pd.set_option('display.max_columns', None)

Using TensorFlow backend.

Since we are going to use big datasets, and we'll need to load them more times, we define a commodity function that deletes all user defined variables, in order to free some memory.

def _del_all():
    %reset_selective -f [^_]

1. Data cleaning¶

Based on the kernel by Hou, Saha, Tsang.¶

We execute the code in recommendation_system_preprocessing.ipynb in order to clean the data and to reduce the size of the dataset, using pickles instead of json and dropping unnecessary columns.

We explore the resulting datasets:

dataset_list = _glob.glob("../dataset/[!checked]*.pickle")
for d in dataset_list:
    dataset = _pd.read_pickle(d)
    
    f = _os.path.splitext(_os.path.basename(d))[0]
    c = ", ".join(list(dataset.columns))
    s = dataset.shape
    
    print("Dataset '" + f + "':")
    print("\tfeatures:", c)
    print("\tshape:", s)
    print()

Dataset 'all_checkin':
	features: business_id, date
	shape: (57402, 2)

Dataset 'all_review':
	features: review_id, user_id, business_id, stars, useful, funny, cool, text, date
	shape: (4201684, 9)

Dataset 'all_tips':
	features: restaurant_name, tips_date, user_id
	shape: (770878, 3)

Dataset 'all_users':
	features: user_id, user_name, average_stars, yelping_since, review, years_of_elite, fans, useful, cool, funny, friends
	shape: (1148098, 11)

Dataset 'restaurants':
	features: name, business_id, address, cuisine, postal_code, latitude, longitude, review_count, stars, OutdoorSeating, BusinessAcceptsCreditCards, RestaurantsDelivery, RestaurantsReservations, WiFi, Alcohol, categories, city, Monday_Open, Tuesday_Open, Wednesday_Open, Thursday_Open, Friday_Open, Saturday_Open, Sunday_Open, Monday_Close, Tuesday_Close, Wednesday_Close, Thursday_Close, Friday_Close, Saturday_Close, Sunday_Close
	shape: (59371, 31)

_del_all()

2. Fake Review Detection¶

Based on Zhiwei Zhang's work and code.¶

Then, in order to filter out deceptive reviews, that could alter the results of our analysis, we load the model based on Support Vector Machine defined in Yelp_sentiment_analysis/Scripts/fake_reviews.ipynb by Zhiwei Zhang, that has the best scores for accuracy, precision, recall and f1-score.

vectorizer = _jl.load('../models/tfidf_vectorizer.joblib')
svc = _jl.load('../models/fake_review_svc_model.joblib')

Now, we can apply this model to our data.

review = _pd.read_pickle("../dataset/all_review.pickle")

review.head()

texts = list(review["text"])
X = vectorizer.transform(texts)
predictions = svc.predict(X)

print(type(predictions))
print("SVC predictions:", predictions)

<class 'numpy.ndarray'>
SVC predictions: [1 1 1 ... 1 1 1]

Now we repeat the whole process with a different model that allows us to obtain real weights instead of a binary evaluation.

cal_svc = _jl.load('../models/fake_review_cal_svc_model.joblib')
cal_predictions = cal_svc.predict_proba(X)

print("Calibrated SVC predictions:\n", cal_predictions)
cal_predictions = _np.array([x[1] for x in cal_predictions])
print("Calibrated SVC predictions for class '1':\n", cal_predictions)

Calibrated SVC predictions:
 [[0.12112422 0.87887578]
 [0.011      0.989     ]
 [0.03538102 0.96461898]
 ...
 [0.36147536 0.63852464]
 [0.00379598 0.99620402]
 [0.00133685 0.99866315]]
Calibrated SVC predictions for class '1':
 [0.87887578 0.989      0.96461898 ... 0.63852464 0.99620402 0.99866315]

print("columns before:\n", review.columns)
checked_review = review.assign(bin_truth_score=predictions, real_truth_score=cal_predictions)
print("columns after:\n", checked_review.columns)

columns before:
 Index(['review_id', 'user_id', 'business_id', 'stars', 'useful', 'funny',
       'cool', 'text', 'date'],
      dtype='object')
columns after:
 Index(['review_id', 'user_id', 'business_id', 'stars', 'useful', 'funny',
       'cool', 'text', 'date', 'bin_truth_score', 'real_truth_score'],
      dtype='object')

Let's see what we just obtained.

checked_review[['review_id', 'text', 'bin_truth_score', 'real_truth_score']].head()

data = checked_review['bin_truth_score']
_plt.hist(data, weights=_np.ones(len(data)) / len(data))
_plt.title("SVC labels distribution")
_plt.gca().yaxis.set_major_formatter(_PercentFormatter(1))
_plt.show()

data = checked_review['real_truth_score']
_plt.hist(data, weights=_np.ones(len(data)) / len(data))
_plt.title("Calibrated SVC labels distribution")
_plt.gca().yaxis.set_major_formatter(_PercentFormatter(1))
_plt.show()

Finally, we can save the new dataset without the text column, in order to save space and computation time.

checked_review.drop(columns=['text'], inplace=True)
checked_review.to_pickle('../dataset/checked_review.pickle')

Check that everything has worked properly.

final_review = _pd.read_pickle('../dataset/checked_review.pickle')
print(final_review.columns)
final_review.head()

Index(['review_id', 'user_id', 'business_id', 'stars', 'useful', 'funny',
       'cool', 'date', 'bin_truth_score', 'real_truth_score'],
      dtype='object')

_del_all()

3. Historical features¶

Following this paper, we add some historical features to our dataset:

user-level features:
1.1. average of the ratings given by a certain user,
1.2. number of reviews written by a certain user,
business-level features:
2.1. average of the ratings given to a certain restaurant,
2.2. number of reviews written about a certain restaurant,
user-business features:
3.1. average rating given by a certain user to each category,
3.2. average of the ratings given by a certain user to the categories of a certain restaurant.

Before proceeding with the computation of the new features, we have to split the dataset in three parts:

Test set, from the last day considered in the dataset, to the previous M months;
Training set, from the day before the beginning of the test set, up to N months before;
History, the remaining part of the dataset, used to compute historical features.

We pick m=3 and n=8, so the test set goes from 9/1/2018 to 11/30/2018, the training set goes from 1/1/2018 to 8/31/2018, the history contains the remaining data, from 10/12/2004 to 12/31/2017.

review_all = _pd.read_pickle("../dataset/checked_review.pickle")
review_test = review_all[review_all['date']>=_np.datetime64('2018-09-01')]
review_train = review_all[(review_all['date']>=_np.datetime64('2018-01-01')) & (review_all['date']<_np.datetime64('2018-09-01'))]
# review_hist = review_all[review_all['date']<_np.datetime64('2018-01-01')]

review_test.to_pickle('../dataset/m2_n9/review_test.pickle')
review_train.to_pickle('../dataset/m2_n9/review_train.pickle')
# review_hist.to_pickle('../dataset/m2_n9/review_hist.pickle')

tips_all = _pd.read_pickle("../dataset/all_tips.pickle")
tips_test = tips_all[tips_all['tips_date']>=_np.datetime64('2018-10-01')]
tips_train = tips_all[(tips_all['tips_date']>=_np.datetime64('2018-01-01')) & (tips_all['tips_date']<_np.datetime64('2018-10-01'))]
tips_hist = tips_all[tips_all['tips_date']<_np.datetime64('2018-01-01')]

tips_test.to_pickle('../dataset/m2_n9/tips_test.pickle')
tips_train.to_pickle('../dataset/m2_n9/tips_train.pickle')
tips_hist.to_pickle('../dataset/m2_n9/tips_hist.pickle')

_del_all()

User-level features¶

review_hist = _pd.read_pickle('../dataset/m2_n9/review_hist.pickle')
users = _pd.read_pickle("../dataset/all_users.pickle")

avg_stars = review_hist['stars'].mean()

users = users.assign(average_stars=avg_stars)
users = users.assign(num_reviews=0)
users = users.assign(average_stars_bin=avg_stars)
users = users.assign(num_reviews_bin=0)
users = users.assign(average_stars_real=avg_stars)
users = users.assign(num_reviews_real=0)
users = users.set_index('user_id')
users.head()

def _f(grouped):
    d = {}
    
    d['num'] = grouped['stars'].size
    d['stars'] = grouped['stars'].mean()
    
    non_fake = _np.ma.masked_where(grouped['bin_truth_score']<=0, grouped['stars']).compressed()
    d['num_bin'] = non_fake.size
    d['stars_bin'] = non_fake.mean()
    
    d['num_real'] = grouped['real_truth_score'].sum()
    d['stars_real'] = _np.average(grouped['stars'], weights=grouped['real_truth_score'])
    
    return _pd.Series(d, index=['num', 'stars', 'num_bin', 'stars_bin', 'num_real', 'stars_real'])

grouped_reviews = review_hist.groupby('user_id').apply(_f)
grouped_reviews.head()

import random
import statistics

current_milli_time = lambda: int(round(_time.time() * 1000))

def get_time(df):
    us_id = random.choice(grouped_reviews.index)
    x = random.randrange(1000)
    t = current_milli_time()
    df.loc[us_id, ["test"]] = x
    t0 = current_milli_time()
    return t0-t

def get_time_mul(df):
    us_id = random.choice(grouped_reviews.index)
    x = random.randrange(1000)
    y = random.randrange(1000)
    z = random.randrange(1000)
    t = current_milli_time()
    df.loc[us_id, ["test", "ciao", "prova"]] = [x, y, z]
    t0 = current_milli_time()
    return t0-t

def test():
    df = users.copy()
    df['test'] = -1
    times = []
    for i in range(1000):
         times += [get_time(df)]
    avg_time = statistics.mean(times)
    del df
    return avg_time

def test_mul():
    df = users.copy()
    df['test'] = -1
    df['ciao'] = -1
    df['prova'] = -1
    times = []
    for i in range(1000):
         times += [get_time(df)]
    avg_time = statistics.mean(times)
    del df
    return avg_time

def tot_time(ops, x, k):
    time_millis = ops * k * x
    hours = time_millis/1000/60/60
    return hours

tot = len(grouped_reviews)
x = test()
print("hours:", tot_time(tot, x, 6))
x = test_mul()
print("hours mul:", tot_time(tot, x, 1))

hours: 53.065662455
hours mul: 8.774814544166667

count = 1
tot = len(grouped_reviews)
print("tot:", tot)

for index, row in grouped_reviews.iterrows():
    uid = index
    num = row['num']
    stars = row['stars']
    num_bin = row['num_bin']
    stars_bin = row['stars_bin']
    num_real = row['num_real']
    stars_real = row['stars_real']
    
    cols = ["num_reviews", "average_stars", "num_reviews_bin",
            "average_stars_bin", "num_reviews_real", "average_stars_real"]
    vals = [num, stars, num_bin, stars_bin, num_real, stars_real]
    users.loc[uid, cols] = vals
    
    count += 1
    if count % 1000 == 0:
        percent = (count/tot)*100
        print("row {}/{} - {}%".format(count, tot, percent))

tot: 954447
row 1000/954447 - 0.10477271131870078%
row 2000/954447 - 0.20954542263740156%
row 3000/954447 - 0.31431813395610236%
row 4000/954447 - 0.4190908452748031%
row 5000/954447 - 0.5238635565935038%
row 6000/954447 - 0.6286362679122047%
row 7000/954447 - 0.7334089792309054%
row 8000/954447 - 0.8381816905496062%
row 9000/954447 - 0.942954401868307%
row 10000/954447 - 1.0477271131870076%
[...]
row 945000/954447 - 99.01021219617223%
row 946000/954447 - 99.11498490749094%
row 947000/954447 - 99.21975761880964%
row 948000/954447 - 99.32453033012834%
row 949000/954447 - 99.42930304144704%
row 950000/954447 - 99.53407575276574%
row 951000/954447 - 99.63884846408445%
row 952000/954447 - 99.74362117540313%
row 953000/954447 - 99.84839388672184%
row 954000/954447 - 99.95316659804054%

users = users.reset_index()
users.to_pickle('../dataset/m2_n9/users.pickle')
_del_all()

Business-level features¶

restaurants = _pd.read_pickle("../dataset/restaurants.pickle")
review_hist = _pd.read_pickle('../dataset/m2_n9/review_hist.pickle')
avg_stars = review_hist['stars'].mean()

restaurants = restaurants.assign(average_stars=avg_stars)
restaurants = restaurants.assign(num_reviews=0)
restaurants = restaurants.assign(average_stars_bin=avg_stars)
restaurants = restaurants.assign(num_reviews_bin=0)
restaurants = restaurants.assign(average_stars_real=avg_stars)
restaurants = restaurants.assign(num_reviews_real=0)
restaurants = restaurants.set_index('business_id')
restaurants.head()

grouped_reviews = review_hist.groupby('business_id').apply(_f)
grouped_reviews.head()

C:\Users\super\Anaconda3\envs\tf-gpu\lib\site-packages\ipykernel_launcher.py:10: RuntimeWarning: Mean of empty slice.
  # Remove the CWD from sys.path while we load stuff.
C:\Users\super\Anaconda3\envs\tf-gpu\lib\site-packages\numpy\core\_methods.py:80: RuntimeWarning: invalid value encountered in double_scalars
  ret = ret.dtype.type(ret / rcount)

count = 1
tot = len(grouped_reviews)
print("tot:", tot)

for index, row in grouped_reviews.iterrows():
    uid = index
    num = row['num']
    stars = row['stars']
    num_bin = row['num_bin']
    stars_bin = row['stars_bin']
    num_real = row['num_real']
    stars_real = row['stars_real']
    
    cols = ["num_reviews", "average_stars", "num_reviews_bin",
            "average_stars_bin", "num_reviews_real", "average_stars_real"]
    vals = [num, stars, num_bin, stars_bin, num_real, stars_real]
    restaurants.loc[uid, cols] = vals
    
    count += 1
    if count % 1000 == 0:
        percent = (count/tot)*100
        print("row {}/{} - {}%".format(count, tot, percent))

tot: 56850
row 1000/56850 - 1.759014951627089%
row 2000/56850 - 3.518029903254178%
row 3000/56850 - 5.277044854881266%
row 4000/56850 - 7.036059806508356%
row 5000/56850 - 8.795074758135444%
row 6000/56850 - 10.554089709762533%
row 7000/56850 - 12.313104661389621%
row 8000/56850 - 14.072119613016712%
row 9000/56850 - 15.8311345646438%
row 10000/56850 - 17.590149516270888%
[...]
row 47000/56850 - 82.67370272647318%
row 48000/56850 - 84.43271767810026%
row 49000/56850 - 86.19173262972735%
row 50000/56850 - 87.95074758135443%
row 51000/56850 - 89.70976253298153%
row 52000/56850 - 91.46877748460862%
row 53000/56850 - 93.22779243623572%
row 54000/56850 - 94.9868073878628%
row 55000/56850 - 96.74582233948989%
row 56000/56850 - 98.50483729111697%

restaurants = restaurants.reset_index()
restaurants.to_pickle('../dataset/m2_n9/restaurants.pickle')
_del_all()

User - Business level features¶

3.1. average rating given by a certain user to each category¶

restaurants = _pd.read_pickle('../dataset/m2_n9/restaurants.pickle')
restaurants.head()

restaurants.columns

Index(['business_id', 'name', 'address', 'cuisine', 'postal_code', 'latitude',
       'longitude', 'review_count', 'stars', 'OutdoorSeating',
       'BusinessAcceptsCreditCards', 'RestaurantsDelivery',
       'RestaurantsReservations', 'WiFi', 'Alcohol', 'categories', 'city',
       'Monday_Open', 'Tuesday_Open', 'Wednesday_Open', 'Thursday_Open',
       'Friday_Open', 'Saturday_Open', 'Sunday_Open', 'Monday_Close',
       'Tuesday_Close', 'Wednesday_Close', 'Thursday_Close', 'Friday_Close',
       'Saturday_Close', 'Sunday_Close', 'average_stars', 'num_reviews',
       'average_stars_bin', 'num_reviews_bin', 'average_stars_real',
       'num_reviews_real'],
      dtype='object')

review_hist = _pd.read_pickle('../dataset/m2_n9/review_hist.pickle')
review_hist.head()

joined_reviews = review_hist.join(restaurants.set_index('business_id'), on = 'business_id', lsuffix='_review', rsuffix='_rest')
joined_reviews.head()

categories = ', '.join(list(restaurants['categories'].unique()))
categories = categories.split(', ')
print(len(categories))

cat = []
for h in categories:
    if h not in cat:
        cat.append(h)
        
print(len(cat))

cuisines = ', '.join(list(restaurants['cuisine'].unique()))
cuisines = cuisines.split(', ')
print(len(cuisines))

_cuisines_unique = []
for cuisine in cuisines:
    if not cuisine in _cuisines_unique:
        _cuisines_unique.append(cuisine)
        
print("Number of cuisines: {0}".format(len(_cuisines_unique)))
print(_cuisines_unique)

173884
761
249
Number of cuisines: 10
['Chinese', 'Japanese', 'Mexican', 'Italian', 'Others', 'American', 'Korean', 'Mediterranean', 'Thai', 'Asian Fusion']

joined_reviews.to_pickle('../dataset/m2_n9/join_restaurants_reviewhist.pickle')

_del_all()

Checkpoint

joined_reviews =  _pd.read_pickle('../dataset/m2_n9/join_restaurants_reviewhist.pickle')
joined_reviews.head()

# joined_reviews = joined_reviews.reset_index()
joined_reviews = joined_reviews[['review_id', 'user_id', 'business_id', 'bin_truth_score', 'real_truth_score', 'cuisine', 'stars_review']]
joined_reviews.head()

#cuisines_unique = ['Chinese', 'Japanese', 'Mexican', 'Italian', 'Others', 'American', 'Korean', 'Mediterranean', 'Thai', 'Asian Fusion']

def each_cuisine_ratings(grouped):
    d = {}
    index = []
    for cuisine in _cuisines_unique:
        cuisine_av = cuisine + "_av"
        cuisine_records = _np.ma.masked_where(~grouped['cuisine'].str.contains(cuisine), grouped['stars_review']).compressed()
        d[cuisine_av] = cuisine_records.mean()
        index.append(cuisine_av)
    # print("cuisine_av done")
        
    for cuisine in _cuisines_unique:
        cuisine_av_bin = cuisine + "_av_bin"
        #non_fake = _np.ma.masked_where(grouped['bin_truth_score'] < 0, grouped).compressed()
        non_fake = grouped[grouped['bin_truth_score'] > 0]
        cuisine_records = _np.ma.masked_where(~non_fake['cuisine'].str.contains(cuisine), non_fake['stars_review']).compressed()
        d[cuisine_av_bin] = cuisine_records.mean()
        index.append(cuisine_av_bin)
    # print("cuisine_av_bin done")
    
    for cuisine in _cuisines_unique:
        cuisine_av_real = cuisine + "_av_real"
        cuisine_records = _np.ma.masked_where(~grouped['cuisine'].str.contains(cuisine), grouped['stars_review']).compressed()
        cuisine_truth_score = _np.ma.masked_where(~grouped['cuisine'].str.contains(cuisine), grouped['real_truth_score']).compressed()
        d[cuisine_av_real] = _np.ma.average(cuisine_records, weights = cuisine_truth_score)
        index.append(cuisine_av_real)
    # print("cuisine_av_real done")
    
    return _pd.Series(d, index = index)

grouped_reviews = joined_reviews.groupby('user_id').apply(each_cuisine_ratings)

C:\Users\super\Anaconda3\envs\tf-gpu\lib\site-packages\ipykernel_launcher.py:8: RuntimeWarning: Mean of empty slice.
  
C:\Users\super\Anaconda3\envs\tf-gpu\lib\site-packages\numpy\core\_methods.py:80: RuntimeWarning: invalid value encountered in double_scalars
  ret = ret.dtype.type(ret / rcount)
C:\Users\super\Anaconda3\envs\tf-gpu\lib\site-packages\ipykernel_launcher.py:17: RuntimeWarning: Mean of empty slice.
C:\Users\super\Anaconda3\envs\tf-gpu\lib\site-packages\numpy\ma\extras.py:607: RuntimeWarning: invalid value encountered in double_scalars
  avg = np.multiply(a, wgt, dtype=result_dtype).sum(axis)/scl

grouped_reviews.head()

Checkpoint 2

users = _pd.read_pickle('../dataset/m2_n9/users.pickle')
users.head()

users = users.assign(av_rat_chinese_cuisine = _np.nan, av_rat_japanese_cuisine = _np.nan, av_rat_mexican_cuisine = _np.nan, 
                     av_rat_italian_cuisine = _np.nan, av_rat_others_cuisine = _np.nan, av_rat_american_cuisine = _np.nan, 
                     av_rat_korean_cuisine = _np.nan, av_rat_mediterranean_cuisine = _np.nan, av_rat_thai_cuisine = _np.nan, 
                     av_rat_asianfusion_cuisine = _np.nan)

users = users.assign(av_rat_chinese_cuisine_bin = _np.nan, av_rat_japanese_cuisine_bin = _np.nan, av_rat_mexican_cuisine_bin = _np.nan, 
                     av_rat_italian_cuisine_bin = _np.nan, av_rat_others_cuisine_bin = _np.nan, av_rat_american_cuisine_bin = _np.nan, 
                     av_rat_korean_cuisine_bin = _np.nan, av_rat_mediterranean_cuisine_bin = _np.nan, av_rat_thai_cuisine_bin = _np.nan, 
                     av_rat_asianfusion_cuisine_bin = _np.nan)

users = users.assign(av_rat_chinese_cuisine_real = _np.nan, av_rat_japanese_cuisine_real = _np.nan, av_rat_mexican_cuisine_real = _np.nan, 
                     av_rat_italian_cuisine_real = _np.nan, av_rat_others_cuisine_real = _np.nan, av_rat_american_cuisine_real = _np.nan, 
                     av_rat_korean_cuisine_real = _np.nan, av_rat_mediterranean_cuisine_real = _np.nan, av_rat_thai_cuisine_real = _np.nan, 
                     av_rat_asianfusion_cuisine_real = _np.nan)

users = users.set_index('user_id')

users.head()

grouped_reviews = _pd.read_pickle('../dataset/m2_n9/grouped_reviews.pickle')
grouped_reviews.head()

# split grouped_reviews and users datasets into n_cores parts, where n_cores is the number of available processors
n_cores = _os.cpu_count()

df_out = _np.array_split(users, n_cores)   # list of input dataframes (from users)

df_out_names = []   # list of paths of output dataframes (from grouped_reviews)
df_in = []
for i, df in enumerate(df_out):
    name = "../dataset/m2_n9/tmp/df_out_" + str(i) + ".pickle"
    df_out_names += [name]
    
    df_tmp = grouped_reviews.loc[df.index]
    df_in += [df_tmp]

C:\Users\super\Anaconda3\envs\tf-gpu\lib\site-packages\ipykernel_launcher.py:12: FutureWarning: 
Passing list-likes to .loc or [] with any missing label will raise
KeyError in the future, you can use .reindex() as an alternative.

See the documentation here:
https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#deprecate-loc-reindex-listlike
  if sys.path[0] == '':

from multiproc_utils import user_business_features

if __name__ ==  '__main__':
    with _Pool(processes=n_cores) as p:
        p.map(user_business_features, zip(df_in, df_out, df_out_names))

users_chunks = []

# add chunks produced by subprocesses
for name in df_out_names:
    df_out_i = _pd.read_pickle(name)
    users_chunks += [df_out_i]
    _os.remove(name)

users = _pd.concat(users_chunks)
users.head()

users = users.reset_index()
users.to_pickle('../dataset/m2_n9/users_2.pickle')

users.shape

(1148098, 46)

users_pre = _pd.read_pickle("../dataset/m2_n9/users.pickle")
users_pre.shape

(1148098, 16)

len(grouped_reviews)

954447

print("expected diff:", users.shape[0]-len(grouped_reviews))

expected diff: 193651

users_tmp = users[['av_rat_chinese_cuisine', 'av_rat_japanese_cuisine', 'av_rat_mexican_cuisine', 'av_rat_italian_cuisine', 
            'av_rat_others_cuisine', 'av_rat_american_cuisine', 'av_rat_korean_cuisine', 'av_rat_mediterranean_cuisine',
            'av_rat_thai_cuisine', 'av_rat_asianfusion_cuisine',
           
           'av_rat_chinese_cuisine_bin', 'av_rat_japanese_cuisine_bin', 'av_rat_mexican_cuisine_bin', 
           'av_rat_italian_cuisine_bin', 'av_rat_others_cuisine_bin', 'av_rat_american_cuisine_bin', 
           'av_rat_korean_cuisine_bin', 'av_rat_mediterranean_cuisine_bin', 'av_rat_thai_cuisine_bin', 
           'av_rat_asianfusion_cuisine_bin',
           
           'av_rat_chinese_cuisine_real', 'av_rat_japanese_cuisine_real', 'av_rat_mexican_cuisine_real', 
           'av_rat_italian_cuisine_real', 'av_rat_others_cuisine_real', 'av_rat_american_cuisine_real', 
           'av_rat_korean_cuisine_real', 'av_rat_mediterranean_cuisine_real', 'av_rat_thai_cuisine_real', 
           'av_rat_asianfusion_cuisine_real']]

count_na = 0
for i, r in users_tmp.iterrows():
        if r.isna().all():
            count_na += 1

print("actual diff:", count_na)

actual diff: 193651

_del_all()

3.2. Average of the ratings given by a certain user to the categories of a certain restaurant.¶

3.2.1 Test set¶

review_test = _pd.read_pickle('../dataset/m2_n9/review_test.pickle')
review_test = review_test.sort_values(by=['review_id'])
review_test = review_test.reset_index(drop = True)
review_test.shape

(153993, 10)

review_test.head()

restaurants = _pd.read_pickle('../dataset/m2_n9/restaurants.pickle')
restaurants = restaurants.reset_index(drop = True)
restaurants = restaurants[['cuisine', 'business_id']]
restaurants.head()

review_test_rest = review_test.join(restaurants.set_index('business_id'), on = 'business_id')
review_test_rest.to_pickle('../dataset/m2_n9/review_test_cuisine.pickle')
review_test_rest.shape

(153993, 11)

review_test_rest.head()

del restaurants

users = _pd.read_pickle('../dataset/m2_n9/users_2.pickle')
users.head()

users = users[['user_id', 'av_rat_chinese_cuisine', 'av_rat_japanese_cuisine', 'av_rat_mexican_cuisine', 'av_rat_italian_cuisine', 
            'av_rat_others_cuisine', 'av_rat_american_cuisine', 'av_rat_korean_cuisine', 'av_rat_mediterranean_cuisine',
            'av_rat_thai_cuisine', 'av_rat_asianfusion_cuisine',
           
           'av_rat_chinese_cuisine_bin', 'av_rat_japanese_cuisine_bin', 'av_rat_mexican_cuisine_bin', 
           'av_rat_italian_cuisine_bin', 'av_rat_others_cuisine_bin', 'av_rat_american_cuisine_bin', 
           'av_rat_korean_cuisine_bin', 'av_rat_mediterranean_cuisine_bin', 'av_rat_thai_cuisine_bin', 
           'av_rat_asianfusion_cuisine_bin',
           
           'av_rat_chinese_cuisine_real', 'av_rat_japanese_cuisine_real', 'av_rat_mexican_cuisine_real', 
           'av_rat_italian_cuisine_real', 'av_rat_others_cuisine_real', 'av_rat_american_cuisine_real', 
           'av_rat_korean_cuisine_real', 'av_rat_mediterranean_cuisine_real', 'av_rat_thai_cuisine_real', 
           'av_rat_asianfusion_cuisine_real']]

users.head()

test_join = review_test_rest.join(users.set_index('user_id'), on = 'user_id', lsuffix = '_test_revirew', rsuffix = '_users')
test_join.shape

(153993, 41)

test_join.head()

test_join.to_pickle('../dataset/m2_n9/join_test_users_review.pickle')
del users, review_test_rest

def _restaturants_users_cuisine_ratings(grouped):
    cuisines = str(grouped['cuisine']).split(", ")
    
    d = {'review_id' : grouped['review_id'],'cuisine_av_hist' : 0, 'cuisine_av_hist_bin' : 0, 'cuisine_av_hist_real': 0}
    index = ['review_id', 'cuisine_av_hist', 'cuisine_av_hist_bin', 'cuisine_av_hist_real']
   
    values = []
    for cuisine in cuisines:
        cui = cuisine.lower().replace(" ", "")
        name = "av_rat_{0}_cuisine".format(cui)
        values.append(grouped[name])
    d['cuisine_av_hist'] = _np.average(values)
    
    values = []
    for cuisine in cuisines:
        cui = cuisine.lower().replace(" ", "")
        name = "av_rat_{0}_cuisine_bin".format(cui)
        values.append(grouped[name])
    d['cuisine_av_hist_bin'] = _np.average(values)
    
    values = []
    for cuisine in cuisines:
        cui = cuisine.lower().replace(" ", "")
        name = "av_rat_{0}_cuisine_real".format(cui)
        values.append(grouped[name])
    d['cuisine_av_hist_real'] = _np.average(values)
    
    return _pd.Series(d, index = index)

applied_test = test_join.apply(_restaturants_users_cuisine_ratings, axis = 1)
applied_test.shape

(153993, 4)

applied_test = applied_test.sort_values(by=['review_id'])
applied_test = applied_test.reset_index(drop = True)
applied_test.head()

applied_test.to_pickle('../dataset/m2_n9/applied_test_users_review.pickle')

review_test.shape

(153993, 10)

review_test.head()

review_test = review_test.assign(cuisine_av_hist = applied_test['cuisine_av_hist'],
                                 cuisine_av_hist_bin = applied_test['cuisine_av_hist_bin'],
                                 cuisine_av_hist_real = applied_test['cuisine_av_hist_real'])
review_test.shape

(153993, 13)

review_test.head()

test_set = review_test
test_set.to_pickle('../dataset/m2_n9/review_test_cuisine_final.pickle')
_del_all()

3.2.2 Training set¶

review_train = _pd.read_pickle('../dataset/m2_n9/review_train.pickle')
review_train = review_train.sort_values(by=['review_id'])
review_train = review_train.reset_index(drop = True)
review_train.shape

(558386, 10)

review_train.head()

restaurants = _pd.read_pickle('../dataset/m2_n9/restaurants.pickle')
restaurants = restaurants.reset_index(drop = True)
restaurants = restaurants[['cuisine', 'business_id']]
restaurants.head()

review_train_rest = review_train.join(restaurants.set_index('business_id'), on = 'business_id')
review_train_rest.to_pickle('../dataset/m2_n9/review_train_cuisine.pickle')
review_train_rest.shape

(558386, 11)

review_train_rest.head()

del restaurants

users = _pd.read_pickle('../dataset/m2_n9/users_2.pickle')
users.head()

users = users[['user_id', 'av_rat_chinese_cuisine', 'av_rat_japanese_cuisine', 'av_rat_mexican_cuisine', 'av_rat_italian_cuisine', 
            'av_rat_others_cuisine', 'av_rat_american_cuisine', 'av_rat_korean_cuisine', 'av_rat_mediterranean_cuisine',
            'av_rat_thai_cuisine', 'av_rat_asianfusion_cuisine',
           
           'av_rat_chinese_cuisine_bin', 'av_rat_japanese_cuisine_bin', 'av_rat_mexican_cuisine_bin', 
           'av_rat_italian_cuisine_bin', 'av_rat_others_cuisine_bin', 'av_rat_american_cuisine_bin', 
           'av_rat_korean_cuisine_bin', 'av_rat_mediterranean_cuisine_bin', 'av_rat_thai_cuisine_bin', 
           'av_rat_asianfusion_cuisine_bin',
           
           'av_rat_chinese_cuisine_real', 'av_rat_japanese_cuisine_real', 'av_rat_mexican_cuisine_real', 
           'av_rat_italian_cuisine_real', 'av_rat_others_cuisine_real', 'av_rat_american_cuisine_real', 
           'av_rat_korean_cuisine_real', 'av_rat_mediterranean_cuisine_real', 'av_rat_thai_cuisine_real', 
           'av_rat_asianfusion_cuisine_real']]

users.head()

train_join = review_train_rest.join(users.set_index('user_id'), on = 'user_id', lsuffix = '_train_revirew', rsuffix = '_users')
train_join.shape

(558386, 41)

train_join.head()

train_join.to_pickle('../dataset/m2_n9/join_train_users_review.pickle')
del users, review_train_rest

applied_train = train_join.apply(_restaturants_users_cuisine_ratings, axis = 1)
applied_train.shape

(558386, 4)

applied_train = applied_train.sort_values(by=['review_id'])
applied_train = applied_train.reset_index(drop = True)
applied_train.head()

applied_train.to_pickle('../dataset/m2_n9/applied_train_users_review.pickle')

review_train.shape

(558386, 10)

review_train.head()

review_train = review_train.assign(cuisine_av_hist = applied_train['cuisine_av_hist'],
                                   cuisine_av_hist_bin = applied_train['cuisine_av_hist_bin'],
                                   cuisine_av_hist_real = applied_train['cuisine_av_hist_real'])
review_train.shape

(558386, 13)

review_train.head()

train_set = review_train
train_set.to_pickle('../dataset/m2_n9/review_train_cuisine_final.pickle')
_del_all()

4. User-based collaborative approach¶

$pred(u, r) = a_u + \frac{\sum_{u_i \in U} sim(u, u_i) * a_{u_i, r} - a_r} {\sum_{u_i \in U} sim(u, u_i)}$

users = _pd.read_pickle('../dataset/m2_n9/users_2.pickle')
users.set_index('user_id', inplace=True)
users.shape

(1148098, 45)

from multiproc_utils import cols_std, cols_bin, cols_real

users = users[[*cols_std, *cols_bin, *cols_real]]
users.shape

(1148098, 30)

users = users.fillna(users.mean())

restaurants = _pd.read_pickle('../dataset/m2_n9/restaurants.pickle')
restaurants.set_index('business_id', inplace=True)

review_hist = _pd.read_pickle('../dataset/m2_n9/review_hist.pickle')
review_hist.shape

(3489305, 10)

Training set¶

review_train = _pd.read_pickle('../dataset/m2_n9/review_train_cuisine_final.pickle')
review_train.set_index('review_id', inplace=True)
review_train.sort_values('business_id', inplace=True)
review_train.shape

(558386, 12)

review_train = review_train.assign(coll_score=_np.nan, coll_score_bin=_np.nan, coll_score_real=_np.nan)
review_train.shape

(558386, 15)

# split review_train, review_hist, restaurants and users datasets into n_cores parts, where n_cores is the number of available processors
n_cores = _os.cpu_count()

review_train_splits = _np.array_split(review_train, n_cores)   # list of input dataframes (from review_train)
review_split_names = []   # list of paths of review_train dataframes
user_splits = []
restaurant_splits = []
review_hist_splits = []

for i, df in enumerate(review_train_splits):
    name = "../dataset/m2_n9/tmp/review_train_split_" + str(i) + ".pickle"
    review_split_names += [name]
    this_review_hist = review_hist.loc[review_hist.business_id.isin(review_hist.business_id.unique())]
    review_hist_splits += [this_review_hist]
    user_ids = set(df.user_id) | set(this_review_hist.user_id)
    user_splits += [users.loc[user_ids]]
    restaurant_splits += [restaurants.loc[df.business_id.unique()]]

from multiproc_utils import set_coll_scores

if __name__ ==  '__main__':
    with _Pool(processes=n_cores) as p:
        p.map(set_coll_scores, zip(review_train_splits, review_split_names, review_hist_splits, user_splits, restaurant_splits))

review_chunks = []

# add chunks produced by subprocesses
for name in review_split_names:
    review_chunks += [_pd.read_pickle(name)]
    _os.remove(name)

review_train = _pd.concat(review_chunks)
review_train.shape

(558386, 15)

review_train.reset_index(inplace=True)
review_train.head()

review_train.to_pickle('../dataset/m2_n9/review_train.pickle')
del review_train

Test set¶

review_test = _pd.read_pickle('../dataset/m2_n9/review_test_cuisine_final.pickle')
review_test.set_index('review_id', inplace=True)
review_test.shape

(153993, 12)

review_test = review_test.assign(coll_score=_np.nan, coll_score_bin=_np.nan, coll_score_real=_np.nan)
review_test.shape

(153993, 15)

# split review_test, review_hist, restaurants and users datasets into n_cores parts, where n_cores is the number of available processors

review_test.sort_values('business_id', inplace=True)
review_test_splits = _np.array_split(review_test, n_cores)   # list of input dataframes (from review_test)
review_split_names = []   # list of paths of review_test dataframes
user_splits = []
restaurant_splits = []
review_hist_splits = []

for i, df in enumerate(review_test_splits):
    name = "../dataset/m2_n9/tmp/review_test_split_" + str(i) + ".pickle"
    review_split_names += [name]
    this_review_hist = review_hist.loc[review_hist.business_id.isin(review_hist.business_id.unique())]
    review_hist_splits += [this_review_hist]
    user_ids = set(df.user_id) | set(this_review_hist.user_id)
    user_splits += [users.loc[user_ids]]
    restaurant_splits += [restaurants.loc[df.business_id.unique()]]

from multiproc_utils import set_coll_scores

if __name__ ==  '__main__':
    with _Pool(processes=n_cores) as p:
        p.map(set_coll_scores, zip(review_test_splits, review_split_names, review_hist_splits, user_splits, restaurant_splits))

review_chunks = []

# add chunks produced by subprocesses
for name in review_split_names:
    review_chunks += [_pd.read_pickle(name)]
    _os.remove(name)

review_test = _pd.concat(review_chunks)
review_test.shape

(153993, 15)

review_test.reset_index(inplace=True)
review_test.head()

review_test.to_pickle('../dataset/m2_n9/review_test.pickle')
_del_all()

5. Some more preprocessing¶

We don't need the dataset checkin, and from the dataset tips we take only the feature "compliments";
The train set is a join of all the data needed for training;
The test set is a join of all the data needed for predictions and performance evaluation (labels included);
The label is a feature 'likes' that is 1 if that user will like that restaurant (4 or 5 stars) or 0 if he/she won't like that restaurant (1, 2 or 3 stars).

5.1 Training set¶

review_train = _pd.read_pickle('../dataset/m2_n9/review_train.pickle')
review_train = review_train.assign(likes = _np.nan)
review_train['likes'] = _np.where(review_train['stars'].isin([4, 5]), 1, 0)
review_train.head(10)

restaurants = _pd.read_pickle('../dataset/m2_n9/restaurants.pickle')
restaurants = restaurants.reset_index(drop = True)
restaurants.head()

review_rest_train = review_train.join(restaurants.set_index('business_id'), on = 'business_id', lsuffix = '_review', rsuffix = '_restaurant')
review_rest_train.head()

print(len(review_train))
print(len(review_rest_train))

558386
558386

tips = _pd.read_pickle('../dataset/m2_n9/tips_train.pickle')
tips = tips.reset_index(drop = True)
tips.head()

tips_agg = tips.groupby(['business_id', 'user_id'])['compliment_count'].agg(_np.sum)
tips_agg.head()

business_id             user_id               
--6MefnULPED_I942VcFNA  EisUuXVeVJN_FcFiE-tqwA    0
--7zmmkVg-IMGaXbuVd0SQ  9LlkZJ7NPsFSFMnIih8X1w    0
--9e1ONYQuAa-CB_Rrw7Tw  2J4PhasBxLtIv-kiS3_FiA    0
                        G2ZiNXL4rZdSxzaDSAAODQ    0
                        GFSZXppCJaO4oiqdgknWnA    0
Name: compliment_count, dtype: int64

review_tip_train = review_rest_train.join(tips_agg, on=['business_id', 'user_id'], lsuffix = '_review', rsuffix = '_tip')
review_tip_train.head()

print(len(review_train))
print(len(review_tip_train))

558386
558386

users = _pd.read_pickle('../dataset/m2_n9/users_2.pickle')
users = users.reset_index(drop = True)
users.head()

train_set = review_tip_train.join(users.set_index('user_id'), on = 'user_id', lsuffix = '_review', rsuffix = '_user')
del review_rest_train, users
train_set.head()

print(len(review_train))
print(len(train_set))

558386
558386

train_set.to_pickle('../dataset/m2_n9/model_train_set.pickle')
_del_all()

5.2 Test set¶

review_test = _pd.read_pickle('../dataset/m2_n9/review_test.pickle')
review_test = review_test.assign(likes = _np.nan)
review_test['likes'] = _np.where(review_test['stars'].isin([4, 5]), 1, 0)
review_test.head()

print(len(review_test))

153993

restaurants = _pd.read_pickle('../dataset/m2_n9/restaurants.pickle')
restaurants = restaurants.reset_index(drop = True)
restaurants.head()

review_rest_test = review_test.join(restaurants.set_index('business_id'), on = 'business_id', lsuffix = '_review', rsuffix = '_restaurant')
del restaurants
review_rest_test.head()

print(len(review_test))
print(len(review_rest_test))

153993
153993

tips = _pd.read_pickle('../dataset/m2_n9/tips_test.pickle')
tips = tips.reset_index(drop = True)
tips.head()

tips_agg = tips.groupby(['business_id', 'user_id'])['compliment_count'].agg(_np.sum)
tips_agg.head()

business_id             user_id               
--9e1ONYQuAa-CB_Rrw7Tw  4vj_0BQeXjCyNB7ESS5mGg    0
                        idlz5ohzqTX5NnOZrxzcsQ    0
                        x_Nu7oNHf4VHwqgV3qLpfg    0
-01XupAWZEXbdNbxNg5mEg  8bt-F30D_tW-ajPG-IpZSA    0
-0DET7VdEQOJVJ_v6klEug  G-6X-llgA_qAxGxocykHzQ    0
Name: compliment_count, dtype: int64

review_tip_test = review_rest_test.join(tips_agg, on=['business_id', 'user_id'], lsuffix = '_review', rsuffix = '_tip')
review_tip_test.head()

print(len(review_test))
print(len(review_tip_test))

153993
153993

users = _pd.read_pickle('../dataset/m2_n9/users_2.pickle')
users = users.reset_index(drop = True)
users.head()

test_set = review_tip_test.join(users.set_index('user_id'), on = 'user_id', lsuffix = '_review', rsuffix = '_user')
del review_rest_test, users
test_set.head()

print(len(review_test))
print(len(test_set))

153993
153993

test_set.to_pickle('../dataset/m2_n9/model_test_set.pickle')
_del_all()

5.3 Prepare data for the models¶

We have to fill missing values in the dataset, and then convert non-numerical features into numerical features, or drop them if they are not necessary for our models, so that the remaining features are readable by our models.

We execute two versions of this step, in order to obtain two different dataset: a complete one and a lighter one. The second one is used for grid search, that duplicates the input dataset to work in parallel and reduce the computation time to a reasonable amount.

5.3.1 Complete version¶

We summarize what kind of data we have at the moment, in order to decide what to do with each feature.

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set.pickle')
train_set.head()

test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set.pickle')
test_set.head()

train_test_set = _pd.concat([train_set, test_set], sort=False)

print("train size:", train_set.shape)
print("test size:", test_set.shape)
print("train_test size:", train_test_set.shape)
print(train_set.shape[0] + test_set.shape[0] == train_test_set.shape[0])
_train_len = train_set.shape[0]

train size: (558386, 99)
test size: (153993, 99)
train_test size: (712379, 99)
True

train_test_set.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 712379 entries, 0 to 153992
Data columns (total 99 columns):
review_id                            712379 non-null object
user_id                              712379 non-null object
business_id                          712379 non-null object
stars_review                         712379 non-null int64
useful_review                        712379 non-null int64
funny_review                         712379 non-null int64
cool_review                          712379 non-null int64
date                                 712379 non-null datetime64[ns]
bin_truth_score                      712379 non-null int64
real_truth_score                     712379 non-null float64
cuisine_av_hist                      269090 non-null float64
cuisine_av_hist_bin                  249642 non-null float64
cuisine_av_hist_real                 269090 non-null float64
coll_score                           269090 non-null float64
coll_score_bin                       249642 non-null float64
coll_score_real                      269090 non-null float64
likes                                712379 non-null int32
name                                 712379 non-null object
address                              712379 non-null object
cuisine                              712379 non-null object
postal_code                          712379 non-null object
latitude                             712379 non-null float64
longitude                            712379 non-null float64
review_count                         712379 non-null int64
stars_restaurant                     712379 non-null float64
OutdoorSeating                       669576 non-null object
BusinessAcceptsCreditCards           588620 non-null object
RestaurantsDelivery                  677719 non-null object
RestaurantsReservations              680010 non-null object
WiFi                                 658240 non-null object
Alcohol                              651883 non-null object
categories                           712379 non-null object
city                                 712379 non-null object
Monday_Open                          645277 non-null object
Tuesday_Open                         668918 non-null object
Wednesday_Open                       679256 non-null object
Thursday_Open                        683039 non-null object
Friday_Open                          683398 non-null object
Saturday_Open                        679080 non-null object
Sunday_Open                          626609 non-null object
Monday_Close                         645277 non-null object
Tuesday_Close                        668918 non-null object
Wednesday_Close                      679256 non-null object
Thursday_Close                       683039 non-null object
Friday_Close                         683398 non-null object
Saturday_Close                       679080 non-null object
Sunday_Close                         626609 non-null object
average_stars_review                 712379 non-null float64
num_reviews_review                   712379 non-null float64
average_stars_bin_review             711373 non-null float64
num_reviews_bin_review               712379 non-null float64
average_stars_real_review            712379 non-null float64
num_reviews_real_review              712379 non-null float64
compliment_count                     17732 non-null float64
user_name                            712379 non-null object
average_stars_user                   712379 non-null float64
yelping_since                        712379 non-null object
review                               712379 non-null int64
years_of_elite                       712379 non-null int64
fans                                 712379 non-null int64
useful_user                          712379 non-null int64
cool_user                            712379 non-null int64
funny_user                           712379 non-null int64
friends                              712379 non-null int64
num_reviews_user                     712379 non-null float64
average_stars_bin_user               694363 non-null float64
num_reviews_bin_user                 712379 non-null float64
average_stars_real_user              712379 non-null float64
num_reviews_real_user                712379 non-null float64
av_rat_chinese_cuisine               140210 non-null float64
av_rat_japanese_cuisine              149309 non-null float64
av_rat_mexican_cuisine               195524 non-null float64
av_rat_italian_cuisine               198832 non-null float64
av_rat_others_cuisine                305838 non-null float64
av_rat_american_cuisine              291851 non-null float64
av_rat_korean_cuisine                73297 non-null float64
av_rat_mediterranean_cuisine         122566 non-null float64
av_rat_thai_cuisine                  105389 non-null float64
av_rat_asianfusion_cuisine           166768 non-null float64
av_rat_chinese_cuisine_bin           127182 non-null float64
av_rat_japanese_cuisine_bin          135034 non-null float64
av_rat_mexican_cuisine_bin           175440 non-null float64
av_rat_italian_cuisine_bin           180287 non-null float64
av_rat_others_cuisine_bin            286214 non-null float64
av_rat_american_cuisine_bin          272580 non-null float64
av_rat_korean_cuisine_bin            66395 non-null float64
av_rat_mediterranean_cuisine_bin     108226 non-null float64
av_rat_thai_cuisine_bin              94816 non-null float64
av_rat_asianfusion_cuisine_bin       151573 non-null float64
av_rat_chinese_cuisine_real          140210 non-null float64
av_rat_japanese_cuisine_real         149309 non-null float64
av_rat_mexican_cuisine_real          195524 non-null float64
av_rat_italian_cuisine_real          198832 non-null float64
av_rat_others_cuisine_real           305838 non-null float64
av_rat_american_cuisine_real         291851 non-null float64
av_rat_korean_cuisine_real           73297 non-null float64
av_rat_mediterranean_cuisine_real    122566 non-null float64
av_rat_thai_cuisine_real             105389 non-null float64
av_rat_asianfusion_cuisine_real      166768 non-null float64
dtypes: datetime64[ns](1), float64(53), int32(1), int64(13), object(31)
memory usage: 540.8+ MB

train_test_types = train_test_set.dtypes

for ind, dtype in train_test_types.iteritems():
    if not _np.issubdtype(dtype, _np.number):
        if "id" not in ind:
            uniq_vals = train_test_set[ind].unique()
            null_vals = train_test_set[ind].isnull().sum()
            print(ind + " - " + str(dtype) + "  - unique: " + str(len(uniq_vals)) + " - nulls: " + str(null_vals))
            print(uniq_vals[:10])
            print()

date - datetime64[ns]  - unique: 699960 - nulls: 0
['2018-01-11T19:55:31.000000000' '2018-02-25T17:47:12.000000000'
 '2018-05-06T04:22:48.000000000' '2018-04-22T17:42:09.000000000'
 '2018-05-21T05:09:07.000000000' '2018-07-08T19:29:51.000000000'
 '2018-08-14T21:59:34.000000000' '2018-08-19T04:01:19.000000000'
 '2018-02-19T03:29:10.000000000' '2018-06-24T18:41:08.000000000']

name - object  - unique: 26090 - nulls: 0
['The Spicy Amigos' "John's Chinese BBQ Restaurant" 'Delmonico Steakhouse'
 'Sunnyside Grill' 'The Bar At Bermuda & St. Rose' 'Mm Mm Pizza' 'Sushiya'
 'Happy Moose Bar and Grill' "Denny's" 'Pio Pio']

address - object  - unique: 35750 - nulls: 0
['821 4 Avenue SW' '328 Highway 7 E, Chalmers Gate 11, Unit 10'
 '3355 Las Vegas Blvd S' '1218 Saint Clair Avenue W' '11624 Bermuda Rd'
 '407 S Central Ave' '1950 Chemin Fer Ã\xa0 Cheval' '9436 State Rte 14'
 '6207 Wilson Mills Rd' '1408 E Blvd']

cuisine - object  - unique: 87 - nulls: 0
['Mexican' 'Chinese' 'Others' 'American' 'Japanese' 'Italian'
 'Asian Fusion' 'Korean' 'Mexican, American' 'Thai']

postal_code - object  - unique: 8309 - nulls: 0
['T2P 0K5' 'L4B 3P7' '89109' 'M6E' '89052' '15317' 'J3E 2T6' '44241'
 '44143' '28203']

OutdoorSeating - object  - unique: 4 - nulls: 42803
['True' 'False' nan 'None']

BusinessAcceptsCreditCards - object  - unique: 4 - nulls: 123759
[nan 'True' 'False' 'None']

RestaurantsDelivery - object  - unique: 4 - nulls: 34660
['False' 'True' nan 'None']

RestaurantsReservations - object  - unique: 4 - nulls: 32369
[nan 'True' 'False' 'None']

WiFi - object  - unique: 4 - nulls: 54139
[nan 'No' 'Free' 'Paid']

Alcohol - object  - unique: 4 - nulls: 60496
['Beer&Wine' 'Full_Bar' 'No' nan]

categories - object  - unique: 22832 - nulls: 0
['Restaurants, Mexican' 'Chinese, Restaurants'
 'Cajun/Creole, Seafood, Steakhouses, Restaurants'
 'Restaurants, Breakfast & Brunch'
 'Nightlife, Beer, Wine & Spirits, Bars, Restaurants, American (New), Food'
 'Restaurants, Pizza, Chicken Wings, Salad'
 'Buffets, Restaurants, Japanese, Sushi Bars'
 'Nightlife, Sports Bars, Restaurants, Bars, American (Traditional)'
 'Breakfast & Brunch, American (Traditional), Restaurants, Diners'
 'Restaurants, Spanish, Peruvian, Colombian, Latin American']

city - object  - unique: 671 - nulls: 0
['Calgary' 'Richmond Hill' 'Las Vegas' 'Toronto' 'Henderson' 'Canonsburg'
 'Sainte-Julie' 'Streetsboro' 'Highland Heights' 'Charlotte']

Monday_Open - object  - unique: 59 - nulls: 67102
[datetime.time(11, 0) datetime.time(17, 0) datetime.time(7, 0)
 datetime.time(0, 0) datetime.time(16, 0) NaT datetime.time(12, 0)
 datetime.time(10, 0) datetime.time(6, 0) datetime.time(6, 30)]

Tuesday_Open - object  - unique: 56 - nulls: 43461
[datetime.time(11, 0) datetime.time(17, 0) datetime.time(7, 0)
 datetime.time(0, 0) datetime.time(16, 0) NaT datetime.time(12, 0)
 datetime.time(10, 0) datetime.time(7, 30) datetime.time(6, 0)]

Wednesday_Open - object  - unique: 60 - nulls: 33123
[datetime.time(11, 0) datetime.time(17, 0) datetime.time(7, 0)
 datetime.time(0, 0) datetime.time(16, 0) NaT datetime.time(12, 0)
 datetime.time(10, 0) datetime.time(7, 30) datetime.time(6, 0)]

Thursday_Open - object  - unique: 62 - nulls: 29340
[datetime.time(11, 0) datetime.time(17, 0) datetime.time(7, 0)
 datetime.time(0, 0) NaT datetime.time(12, 0) datetime.time(10, 0)
 datetime.time(7, 30) datetime.time(6, 0) datetime.time(15, 0)]

Saturday_Open - object  - unique: 59 - nulls: 33299
[datetime.time(11, 0) datetime.time(17, 0) datetime.time(7, 0)
 datetime.time(0, 0) datetime.time(16, 0) NaT datetime.time(12, 0)
 datetime.time(10, 0) datetime.time(7, 30) datetime.time(6, 0)]

Sunday_Open - object  - unique: 62 - nulls: 85770
[NaT datetime.time(11, 0) datetime.time(17, 0) datetime.time(7, 0)
 datetime.time(0, 0) datetime.time(4, 30) datetime.time(12, 0)
 datetime.time(10, 0) datetime.time(7, 30) datetime.time(6, 0)]

Monday_Close - object  - unique: 69 - nulls: 67102
[datetime.time(20, 0) datetime.time(22, 30) datetime.time(22, 0)
 datetime.time(15, 0) datetime.time(0, 0) datetime.time(23, 0)
 datetime.time(21, 0) datetime.time(2, 30) NaT datetime.time(21, 30)]

Tuesday_Close - object  - unique: 70 - nulls: 43461
[datetime.time(20, 0) datetime.time(22, 30) datetime.time(22, 0)
 datetime.time(15, 0) datetime.time(0, 0) datetime.time(23, 0)
 datetime.time(21, 0) datetime.time(2, 30) NaT datetime.time(1, 0)]

Wednesday_Close - object  - unique: 70 - nulls: 33123
[datetime.time(20, 0) datetime.time(22, 30) datetime.time(22, 0)
 datetime.time(15, 0) datetime.time(0, 0) datetime.time(23, 0)
 datetime.time(21, 0) datetime.time(2, 30) NaT datetime.time(1, 0)]

Thursday_Close - object  - unique: 77 - nulls: 29340
[datetime.time(20, 0) datetime.time(22, 30) datetime.time(22, 0)
 datetime.time(15, 0) datetime.time(0, 0) datetime.time(23, 0)
 datetime.time(21, 0) datetime.time(2, 30) NaT datetime.time(1, 0)]

Saturday_Close - object  - unique: 76 - nulls: 33299
[datetime.time(4, 0) datetime.time(22, 30) datetime.time(16, 0)
 datetime.time(0, 0) datetime.time(1, 0) datetime.time(22, 0)
 datetime.time(2, 30) NaT datetime.time(2, 0) datetime.time(21, 30)]

Sunday_Close - object  - unique: 71 - nulls: 85770
[NaT datetime.time(22, 30) datetime.time(22, 0) datetime.time(16, 0)
 datetime.time(0, 0) datetime.time(23, 0) datetime.time(21, 0)
 datetime.time(2, 30) datetime.time(21, 30) datetime.time(15, 0)]

user_name - object  - unique: 41141 - nulls: 0
['Jexie' 'Alex' 'Mermaid' 'Jen' 'Kit' 'Y' 'Belinda' 'Jim' 'Marko' 'Tien']

yelping_since - object  - unique: 4618 - nulls: 0
[datetime.date(2009, 6, 24) datetime.date(2015, 7, 27)
 datetime.date(2012, 1, 1) datetime.date(2013, 2, 27)
 datetime.date(2008, 11, 15) datetime.date(2011, 4, 16)
 datetime.date(2013, 6, 25) datetime.date(2011, 6, 7)
 datetime.date(2011, 6, 4) datetime.date(2016, 7, 5)]

Drop useless features

train_test_set.drop(columns=['date', 'name', 'address', 'yelping_since', 'user_name', 'cuisine'], inplace=True)

Fill missing values

train_test_set['OutdoorSeating'] = train_test_set['OutdoorSeating'].fillna('None')
train_test_set['BusinessAcceptsCreditCards'] = train_test_set['BusinessAcceptsCreditCards'].fillna('None')
train_test_set['RestaurantsDelivery'] = train_test_set['RestaurantsDelivery'].fillna('None')
train_test_set['RestaurantsReservations'] = train_test_set['RestaurantsReservations'].fillna('None')
train_test_set['WiFi'] = train_test_set['WiFi'].fillna('None')
train_test_set['Alcohol'] = train_test_set['Alcohol'].fillna('None')

train_test_set['Monday_Open'] = train_test_set["Monday_Open"].astype(str)
train_test_set['Monday_Open'] = train_test_set['Monday_Open'].fillna(train_test_set['Monday_Open'].mode())
train_test_set['Tuesday_Open'] = train_test_set["Tuesday_Open"].astype(str)
train_test_set['Tuesday_Open'] = train_test_set['Tuesday_Open'].fillna(train_test_set['Tuesday_Open'].mode())
train_test_set['Wednesday_Open'] = train_test_set["Wednesday_Open"].astype(str)
train_test_set['Wednesday_Open'] = train_test_set['Wednesday_Open'].fillna(train_test_set['Wednesday_Open'].mode())
train_test_set['Thursday_Open'] = train_test_set["Thursday_Open"].astype(str)
train_test_set['Thursday_Open'] = train_test_set['Thursday_Open'].fillna(train_test_set['Thursday_Open'].mode())
train_test_set['Friday_Open'] = train_test_set["Friday_Open"].astype(str)
train_test_set['Friday_Open'] = train_test_set['Friday_Open'].fillna(train_test_set['Friday_Open'].mode())
train_test_set['Saturday_Open'] = train_test_set["Saturday_Open"].astype(str)
train_test_set['Saturday_Open'] = train_test_set['Saturday_Open'].fillna(train_test_set['Saturday_Open'].mode())
train_test_set['Sunday_Open'] = train_test_set["Sunday_Open"].astype(str)
train_test_set['Sunday_Open'] = train_test_set['Sunday_Open'].fillna(train_test_set['Sunday_Open'].mode())
train_test_set['Monday_Close'] = train_test_set["Monday_Close"].astype(str)
train_test_set['Monday_Close'] = train_test_set['Monday_Close'].fillna(train_test_set['Monday_Close'].mode())
train_test_set['Tuesday_Close'] = train_test_set["Tuesday_Close"].astype(str)
train_test_set['Tuesday_Close'] = train_test_set['Tuesday_Close'].fillna(train_test_set['Tuesday_Close'].mode())
train_test_set['Wednesday_Close'] = train_test_set["Wednesday_Close"].astype(str)
train_test_set['Wednesday_Close'] = train_test_set['Wednesday_Close'].fillna(train_test_set['Wednesday_Close'].mode())
train_test_set['Thursday_Close'] = train_test_set["Thursday_Close"].astype(str)
train_test_set['Thursday_Close'] = train_test_set['Thursday_Close'].fillna(train_test_set['Thursday_Close'].mode())
train_test_set['Friday_Close'] = train_test_set["Friday_Close"].astype(str)
train_test_set['Friday_Close'] = train_test_set['Friday_Close'].fillna(train_test_set['Friday_Close'].mode())
train_test_set['Saturday_Close'] = train_test_set["Saturday_Close"].astype(str)
train_test_set['Saturday_Close'] = train_test_set['Saturday_Close'].fillna(train_test_set['Saturday_Close'].mode())
train_test_set['Sunday_Close'] = train_test_set["Sunday_Close"].astype(str)
train_test_set['Sunday_Close'] = train_test_set['Sunday_Close'].fillna(train_test_set['Sunday_Close'].mode())

for ind, dtype in train_test_types.iteritems():
    if _np.issubdtype(dtype, _np.floating):
        train_test_set[ind] = train_test_set[ind].fillna(train_test_set[ind].mean())
    elif _np.issubdtype(dtype, _np.integer):
        train_test_set[ind] = train_test_set[ind].fillna(round(train_test_set[ind].mean()))

# check if any feature still has null values
train_test_set.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 712379 entries, 0 to 153992
Data columns (total 93 columns):
review_id                            712379 non-null object
user_id                              712379 non-null object
business_id                          712379 non-null object
stars_review                         712379 non-null int64
useful_review                        712379 non-null int64
funny_review                         712379 non-null int64
cool_review                          712379 non-null int64
bin_truth_score                      712379 non-null int64
real_truth_score                     712379 non-null float64
cuisine_av_hist                      712379 non-null float64
cuisine_av_hist_bin                  712379 non-null float64
cuisine_av_hist_real                 712379 non-null float64
coll_score                           712379 non-null float64
coll_score_bin                       712379 non-null float64
coll_score_real                      712379 non-null float64
likes                                712379 non-null int32
postal_code                          712379 non-null object
latitude                             712379 non-null float64
longitude                            712379 non-null float64
review_count                         712379 non-null int64
stars_restaurant                     712379 non-null float64
OutdoorSeating                       712379 non-null object
BusinessAcceptsCreditCards           712379 non-null object
RestaurantsDelivery                  712379 non-null object
RestaurantsReservations              712379 non-null object
WiFi                                 712379 non-null object
Alcohol                              712379 non-null object
categories                           712379 non-null object
city                                 712379 non-null object
Monday_Open                          712379 non-null object
Tuesday_Open                         712379 non-null object
Wednesday_Open                       712379 non-null object
Thursday_Open                        712379 non-null object
Friday_Open                          712379 non-null object
Saturday_Open                        712379 non-null object
Sunday_Open                          712379 non-null object
Monday_Close                         712379 non-null object
Tuesday_Close                        712379 non-null object
Wednesday_Close                      712379 non-null object
Thursday_Close                       712379 non-null object
Friday_Close                         712379 non-null object
Saturday_Close                       712379 non-null object
Sunday_Close                         712379 non-null object
average_stars_review                 712379 non-null float64
num_reviews_review                   712379 non-null float64
average_stars_bin_review             712379 non-null float64
num_reviews_bin_review               712379 non-null float64
average_stars_real_review            712379 non-null float64
num_reviews_real_review              712379 non-null float64
compliment_count                     712379 non-null float64
average_stars_user                   712379 non-null float64
review                               712379 non-null int64
years_of_elite                       712379 non-null int64
fans                                 712379 non-null int64
useful_user                          712379 non-null int64
cool_user                            712379 non-null int64
funny_user                           712379 non-null int64
friends                              712379 non-null int64
num_reviews_user                     712379 non-null float64
average_stars_bin_user               712379 non-null float64
num_reviews_bin_user                 712379 non-null float64
average_stars_real_user              712379 non-null float64
num_reviews_real_user                712379 non-null float64
av_rat_chinese_cuisine               712379 non-null float64
av_rat_japanese_cuisine              712379 non-null float64
av_rat_mexican_cuisine               712379 non-null float64
av_rat_italian_cuisine               712379 non-null float64
av_rat_others_cuisine                712379 non-null float64
av_rat_american_cuisine              712379 non-null float64
av_rat_korean_cuisine                712379 non-null float64
av_rat_mediterranean_cuisine         712379 non-null float64
av_rat_thai_cuisine                  712379 non-null float64
av_rat_asianfusion_cuisine           712379 non-null float64
av_rat_chinese_cuisine_bin           712379 non-null float64
av_rat_japanese_cuisine_bin          712379 non-null float64
av_rat_mexican_cuisine_bin           712379 non-null float64
av_rat_italian_cuisine_bin           712379 non-null float64
av_rat_others_cuisine_bin            712379 non-null float64
av_rat_american_cuisine_bin          712379 non-null float64
av_rat_korean_cuisine_bin            712379 non-null float64
av_rat_mediterranean_cuisine_bin     712379 non-null float64
av_rat_thai_cuisine_bin              712379 non-null float64
av_rat_asianfusion_cuisine_bin       712379 non-null float64
av_rat_chinese_cuisine_real          712379 non-null float64
av_rat_japanese_cuisine_real         712379 non-null float64
av_rat_mexican_cuisine_real          712379 non-null float64
av_rat_italian_cuisine_real          712379 non-null float64
av_rat_others_cuisine_real           712379 non-null float64
av_rat_american_cuisine_real         712379 non-null float64
av_rat_korean_cuisine_real           712379 non-null float64
av_rat_mediterranean_cuisine_real    712379 non-null float64
av_rat_thai_cuisine_real             712379 non-null float64
av_rat_asianfusion_cuisine_real      712379 non-null float64
dtypes: float64(53), int32(1), int64(13), object(26)
memory usage: 508.2+ MB

Convert non-numerical features

train_test_set.shape

(712379, 93)

cat_cols = ['OutdoorSeating', 'BusinessAcceptsCreditCards', 'RestaurantsDelivery', 'RestaurantsReservations', 'WiFi',
            'Alcohol', 'city', 'Monday_Open', 'Tuesday_Open', 'Wednesday_Open', 'Thursday_Open', 'Friday_Open',
            'Saturday_Open', 'Sunday_Open', 'Monday_Close', 'Tuesday_Close', 'Wednesday_Close', 'Thursday_Close',
            'Friday_Close', 'Saturday_Close', 'Sunday_Close']
train_test_set = _pd.get_dummies(train_test_set, columns=cat_cols, prefix=cat_cols)
train_test_set.shape

(712379, 1689)

oe = _OrdinalEncoder()

train_test_set['postal_code'] = oe.fit_transform(train_test_set['postal_code'].to_numpy().reshape(-1, 1))
train_test_set.shape

(712379, 1689)

categories = train_test_set['categories'].str.get_dummies(',')
f1 = lambda x: "categories_" + x
categories.rename(columns=f1, inplace=True)
train_test_set[categories.columns] = categories
train_test_set.drop(columns=['categories'], inplace=True)
train_test_set.shape

(712379, 2774)

The resulting dataset

train_test_set.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 712379 entries, 0 to 153992
Columns: 2774 entries, review_id to categories_Yoga
dtypes: float64(54), int32(1), int64(1099), object(3), uint8(1617)
memory usage: 7.2+ GB

train_set = train_test_set[:_train_len]
test_set = train_test_set[_train_len:]

train_set.head(10)

train_set.shape

(558386, 2774)

train_set.to_pickle('../dataset/m2_n9/model_train_set_2.pickle')

test_set.head(10)

test_set.shape

(153993, 2774)

test_set.to_pickle('../dataset/m2_n9/model_test_set_2.pickle')

_del_all()

5.3.2 Dimensionality reduction version¶

We summarize what kind of data we have at the moment, in order to decide what to do with each feature.

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set.pickle')
train_set.head()

test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set.pickle')
test_set.head()

train_test_set = _pd.concat([train_set, test_set], sort=False)

print("train size:", train_set.shape)
print("test size:", test_set.shape)
print("train_test size:", train_test_set.shape)
print(train_set.shape[0] + test_set.shape[0] == train_test_set.shape[0])
_train_len = train_set.shape[0]

train size: (558386, 99)
test size: (153993, 99)
train_test size: (712379, 99)
True

Drop useless features

train_test_set.drop(columns=['date', 'name', 'address', 'yelping_since', 'user_name', 'cuisine'], inplace=True)

Fill missing values

train_test_set['OutdoorSeating'] = train_test_set['OutdoorSeating'].fillna('None')
train_test_set['BusinessAcceptsCreditCards'] = train_test_set['BusinessAcceptsCreditCards'].fillna('None')
train_test_set['RestaurantsDelivery'] = train_test_set['RestaurantsDelivery'].fillna('None')
train_test_set['RestaurantsReservations'] = train_test_set['RestaurantsReservations'].fillna('None')
train_test_set['WiFi'] = train_test_set['WiFi'].fillna('None')
train_test_set['Alcohol'] = train_test_set['Alcohol'].fillna('None')

train_test_set['Monday_Open'] = train_test_set["Monday_Open"].astype(str)
train_test_set['Monday_Open'] = train_test_set['Monday_Open'].fillna(train_test_set['Monday_Open'].mode())
train_test_set['Tuesday_Open'] = train_test_set["Tuesday_Open"].astype(str)
train_test_set['Tuesday_Open'] = train_test_set['Tuesday_Open'].fillna(train_test_set['Tuesday_Open'].mode())
train_test_set['Wednesday_Open'] = train_test_set["Wednesday_Open"].astype(str)
train_test_set['Wednesday_Open'] = train_test_set['Wednesday_Open'].fillna(train_test_set['Wednesday_Open'].mode())
train_test_set['Thursday_Open'] = train_test_set["Thursday_Open"].astype(str)
train_test_set['Thursday_Open'] = train_test_set['Thursday_Open'].fillna(train_test_set['Thursday_Open'].mode())
train_test_set['Friday_Open'] = train_test_set["Friday_Open"].astype(str)
train_test_set['Friday_Open'] = train_test_set['Friday_Open'].fillna(train_test_set['Friday_Open'].mode())
train_test_set['Saturday_Open'] = train_test_set["Saturday_Open"].astype(str)
train_test_set['Saturday_Open'] = train_test_set['Saturday_Open'].fillna(train_test_set['Saturday_Open'].mode())
train_test_set['Sunday_Open'] = train_test_set["Sunday_Open"].astype(str)
train_test_set['Sunday_Open'] = train_test_set['Sunday_Open'].fillna(train_test_set['Sunday_Open'].mode())
train_test_set['Monday_Close'] = train_test_set["Monday_Close"].astype(str)
train_test_set['Monday_Close'] = train_test_set['Monday_Close'].fillna(train_test_set['Monday_Close'].mode())
train_test_set['Tuesday_Close'] = train_test_set["Tuesday_Close"].astype(str)
train_test_set['Tuesday_Close'] = train_test_set['Tuesday_Close'].fillna(train_test_set['Tuesday_Close'].mode())
train_test_set['Wednesday_Close'] = train_test_set["Wednesday_Close"].astype(str)
train_test_set['Wednesday_Close'] = train_test_set['Wednesday_Close'].fillna(train_test_set['Wednesday_Close'].mode())
train_test_set['Thursday_Close'] = train_test_set["Thursday_Close"].astype(str)
train_test_set['Thursday_Close'] = train_test_set['Thursday_Close'].fillna(train_test_set['Thursday_Close'].mode())
train_test_set['Friday_Close'] = train_test_set["Friday_Close"].astype(str)
train_test_set['Friday_Close'] = train_test_set['Friday_Close'].fillna(train_test_set['Friday_Close'].mode())
train_test_set['Saturday_Close'] = train_test_set["Saturday_Close"].astype(str)
train_test_set['Saturday_Close'] = train_test_set['Saturday_Close'].fillna(train_test_set['Saturday_Close'].mode())
train_test_set['Sunday_Close'] = train_test_set["Sunday_Close"].astype(str)
train_test_set['Sunday_Close'] = train_test_set['Sunday_Close'].fillna(train_test_set['Sunday_Close'].mode())

for ind, dtype in train_test_types.iteritems():
    if _np.issubdtype(dtype, _np.floating):
        train_test_set[ind] = train_test_set[ind].fillna(train_test_set[ind].mean())
    elif _np.issubdtype(dtype, _np.integer):
        train_test_set[ind] = train_test_set[ind].fillna(round(train_test_set[ind].mean()))

# check any feature still has null values
train_test_set.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 712379 entries, 0 to 153992
Data columns (total 93 columns):
review_id                            712379 non-null object
user_id                              712379 non-null object
business_id                          712379 non-null object
stars_review                         712379 non-null int64
useful_review                        712379 non-null int64
funny_review                         712379 non-null int64
cool_review                          712379 non-null int64
bin_truth_score                      712379 non-null int64
real_truth_score                     712379 non-null float64
cuisine_av_hist                      712379 non-null float64
cuisine_av_hist_bin                  712379 non-null float64
cuisine_av_hist_real                 712379 non-null float64
coll_score                           712379 non-null float64
coll_score_bin                       712379 non-null float64
coll_score_real                      712379 non-null float64
likes                                712379 non-null int32
postal_code                          712379 non-null object
latitude                             712379 non-null float64
longitude                            712379 non-null float64
review_count                         712379 non-null int64
stars_restaurant                     712379 non-null float64
OutdoorSeating                       712379 non-null object
BusinessAcceptsCreditCards           712379 non-null object
RestaurantsDelivery                  712379 non-null object
RestaurantsReservations              712379 non-null object
WiFi                                 712379 non-null object
Alcohol                              712379 non-null object
categories                           712379 non-null object
city                                 712379 non-null object
Monday_Open                          712379 non-null object
Tuesday_Open                         712379 non-null object
Wednesday_Open                       712379 non-null object
Thursday_Open                        712379 non-null object
Friday_Open                          712379 non-null object
Saturday_Open                        712379 non-null object
Sunday_Open                          712379 non-null object
Monday_Close                         712379 non-null object
Tuesday_Close                        712379 non-null object
Wednesday_Close                      712379 non-null object
Thursday_Close                       712379 non-null object
Friday_Close                         712379 non-null object
Saturday_Close                       712379 non-null object
Sunday_Close                         712379 non-null object
average_stars_review                 712379 non-null float64
num_reviews_review                   712379 non-null float64
average_stars_bin_review             712379 non-null float64
num_reviews_bin_review               712379 non-null float64
average_stars_real_review            712379 non-null float64
num_reviews_real_review              712379 non-null float64
compliment_count                     712379 non-null float64
average_stars_user                   712379 non-null float64
review                               712379 non-null int64
years_of_elite                       712379 non-null int64
fans                                 712379 non-null int64
useful_user                          712379 non-null int64
cool_user                            712379 non-null int64
funny_user                           712379 non-null int64
friends                              712379 non-null int64
num_reviews_user                     712379 non-null float64
average_stars_bin_user               712379 non-null float64
num_reviews_bin_user                 712379 non-null float64
average_stars_real_user              712379 non-null float64
num_reviews_real_user                712379 non-null float64
av_rat_chinese_cuisine               712379 non-null float64
av_rat_japanese_cuisine              712379 non-null float64
av_rat_mexican_cuisine               712379 non-null float64
av_rat_italian_cuisine               712379 non-null float64
av_rat_others_cuisine                712379 non-null float64
av_rat_american_cuisine              712379 non-null float64
av_rat_korean_cuisine                712379 non-null float64
av_rat_mediterranean_cuisine         712379 non-null float64
av_rat_thai_cuisine                  712379 non-null float64
av_rat_asianfusion_cuisine           712379 non-null float64
av_rat_chinese_cuisine_bin           712379 non-null float64
av_rat_japanese_cuisine_bin          712379 non-null float64
av_rat_mexican_cuisine_bin           712379 non-null float64
av_rat_italian_cuisine_bin           712379 non-null float64
av_rat_others_cuisine_bin            712379 non-null float64
av_rat_american_cuisine_bin          712379 non-null float64
av_rat_korean_cuisine_bin            712379 non-null float64
av_rat_mediterranean_cuisine_bin     712379 non-null float64
av_rat_thai_cuisine_bin              712379 non-null float64
av_rat_asianfusion_cuisine_bin       712379 non-null float64
av_rat_chinese_cuisine_real          712379 non-null float64
av_rat_japanese_cuisine_real         712379 non-null float64
av_rat_mexican_cuisine_real          712379 non-null float64
av_rat_italian_cuisine_real          712379 non-null float64
av_rat_others_cuisine_real           712379 non-null float64
av_rat_american_cuisine_real         712379 non-null float64
av_rat_korean_cuisine_real           712379 non-null float64
av_rat_mediterranean_cuisine_real    712379 non-null float64
av_rat_thai_cuisine_real             712379 non-null float64
av_rat_asianfusion_cuisine_real      712379 non-null float64
dtypes: float64(53), int32(1), int64(13), object(26)
memory usage: 508.2+ MB

Convert non-numerical features

We print and plot the distribution of the cities, to see the long tail and decide how many of them to keep.

city_count = train_test_set['city'].value_counts()
print(city_count.to_string())
print(city_count.shape)

Las Vegas                           208437
Phoenix                              71126
Toronto                              57047
Charlotte                            40190
Scottsdale                           36579
Pittsburgh                           26891
Henderson                            21518
MontrÃ©al                            18531
Tempe                                18354
Mesa                                 17235
Chandler                             15441
Gilbert                              14055
Cleveland                            12873
Glendale                             10742
Madison                               9688
Markham                               8248
Calgary                               7958
Mississauga                           7192
Peoria                                6582
North Las Vegas                       4771
Surprise                              4117
Richmond Hill                         3886
Goodyear                              3020
Concord                               2905
Scarborough                           2544
Champaign                             2513
Vaughan                               2496
Avondale                              2451
Huntersville                          2046
Lakewood                              2008
Matthews                              1986
North York                            1977
Fort Mill                             1851
Brampton                              1643
Cave Creek                            1538
Gastonia                              1282
Cornelius                             1278
Etobicoke                             1229
Akron                                 1188
Westlake                              1118
Boulder City                          1059
Thornhill                              985
Strongsville                           957
Middleton                              930
Mentor                                 864
Fountain Hills                         836
Cuyahoga Falls                         828
Rock Hill                              820
Cleveland Heights                      813
Parma                                  789
Kent                                   745
Litchfield Park                        745
Aurora                                 712
North Olmsted                          693
Beachwood                              654
Independence                           648
Pineville                              647
Medina                                 647
Newmarket                              637
Belmont                                631
Monroeville                            610
Willoughby                             607
Rocky River                            579
Solon                                  576
Oakville                               573
Indian Trail                           545
Avon                                   543
Bridgeville                            512
Davidson                               507
Hudson                                 500
Pickering                              493
Laval                                  491
Urbana                                 491
Ajax                                   479
Fitchburg                              464
Waxhaw                                 462
Montreal                               441
Sun Prairie                            434
Homestead                              429
Paradise Valley                        425
Chagrin Falls                          410
Stow                                   406
Canonsburg                             400
Woodbridge                             374
Wexford                                368
Whitby                                 367
Woodmere                               360
Carnegie                               358
Lyndhurst                              358
Coraopolis                             341
Fairlawn                               341
Bethel Park                            331
Unionville                             326
Elyria                                 319
Verona                                 315
Berea                                  313
Spring Valley                          308
Sewickley                              306
Orange Village                         297
Carefree                               294
Denver                                 288
Brossard                               284
Brunswick                              282
Sun City                               280
Indian Land                            267
Broadview Heights                      259
Fairview Park                          254
Monona                                 253
Laveen                                 253
Harrisburg                             252
Avon Lake                              249
Tolleson                               243
Twinsburg                              243
Mayfield Heights                       242
Buckeye                                242
Middleburg Heights                     239
Macedonia                              238
Dollard-des-Ormeaux                    238
Brooklyn                               233
Oakmont                                224
Mint Hill                              224
Painesville                            223
North Royalton                         210
Kannapolis                             204
Amherst                                203
Queen Creek                            203
Streetsboro                            196
Monroe                                 194
Clover                                 193
El Mirage                              192
Orange                                 188
Murrysville                            188
Mount Holly                            185
Verdun                                 170
Dorval                                 168
Maple                                  167
Lake Wylie                             167
Chardon                                166
University Heights                     165
Valley View                            153
Highland Heights                       153
South Las Vegas                        153
Airdrie                                150
York                                   149
Tega Cay                               148
Lorain                                 147
West Mifflin                           146
McMurray                               144
Moon Township                          144
Chesterland                            141
Euclid                                 141
Allison Park                           140
Irwin                                  139
Brecksville                            138
Westmount                              138
Waunakee                               137
Brook Park                             137
Stoughton                              134
Pointe-Claire                          130
Northfield                             127
Warrensville Heights                   127
Saint-Laurent                          126
South Euclid                           126
East York                              124
McKees Rocks                           120
Bellevue                               117
Olmsted Falls                          116
Braddock                               116
Blue Diamond                           109
Ahwatukee                              109
New Kensington                         108
Seven Hills                            107
Mooresville                            106
Upper Saint Clair                      105
Longueuil                              105
North Ridgeville                       102
West Homestead                          98
Eastlake                                97
Oregon                                  97
Gibsonia                                96
Sun City West                           92
Anthem                                  90
Willoughby Hills                        90
Pleasant Hills                          90
Youngtown                               88
Willowick                               88
DeForest                                86
Parma Heights                           84
Upper St Clair                          83
Tallmadge                               82
Stouffville                             81
Bedford                                 80
Las  Vegas                              79
Robinson Township                       79
Whitchurch-Stouffville                  77
LasVegas                                77
Peninsula                               76
Saint-Sauveur                           75
Stallings                               74
Bolton                                  73
Trafford                                72
Glenshaw                                72
Mount Horeb                             71
Wesley Chapel                           70
Moreland Hills                          70
Pierrefonds                             69
Missisauga                              68
Higley                                  67
Heidelberg                              67
Millvale                                66
Vaudreuil-Dorion                        65
Lachine                                 64
N. Las Vegas                            64
Wickliffe                               62
Ambridge                                62
Finleyville                             62
Burton                                  61
Munroe Falls                            61
Brentwood                               59
Shaker Heights                          59
Tarentum                                59
McKeesport                              59
Sun Lakes                               58
Georgetown                              57
Rantoul                                 57
Bay Village                             57
East Gwillimbury                        56
Cecil                                   56
Wadsworth                               56
Lasalle                                 56
Richfield                               55
Mt. Lebanon                             55
Kirkland                                55
Mc Kees Rocks                           54
Oakdale                                 54
North Versailles                        54
Mayfield                                53
MESA                                    53
King City                               52
Blakeney                                51
Enterprise                              50
Ross Township                           50
Cramerton                               50
Mahomet                                 49
North Randall                           49
Pepper Pike                             48
McAdenville                             48
Cottage Grove                           47
Wilkinsburg                             47
Outremont                               47
lyndhurst                               46
Savoy                                   45
Grafton                                 45
Turtle Creek                            45
Maple Heights                           45
Lowell                                  44
Springdale                              44
Concord Mills                           44
Valley City                             43
Richmond Heights                        43
White Oak                               43
Bradford                                43
Bedford Heights                         42
West View                               42
Dallas                                  42
GILBERT                                 42
Moon                                    41
L'ÃŽle-Perrot                           40
Grand River                             40
Monticello                              40
Sheffield Village                       40
Hampton Township                        39
Rexdale                                 39
Lower Lawrenceville                     39
Guadalupe                               38
McFarland                               38
Bradford West Gwillimbury               37
North Huntingdon                        37
Copley                                  36
Caledon                                 36
Laveen Village                          36
Summerlin                               35
Fairport Harbor                         35
Gates Mills                             35
Elizabeth                               34
Mentor-on-the-Lake                      34
Saint-Hubert                            33
Mayfield Village                        33
Boucherville                            33
De Forest                               33
Apache Junction                         33
Windsor                                 33
Kirtland                                33
Locust                                  33
Munhall                                 32
Penn Hills                              32
Saint-Lambert                           31
Leetsdale                               31
Harmarville                             31
Tremont                                 31
Aspinwall                               31
Norton                                  30
Tornto                                  30
Crafton                                 29
Garfield Heights                        29
Robinson                                28
Weddington                              27
Sainte-Anne-de-Bellevue                 27
Saint-LÃ©onard                          27
Plum                                    26
Woodmere Village                        26
Blawnox                                 26
La Prairie                              26
Castle Shannon                          26
Rio Verde                               26
Fort McDowell                           25
Middlefield                             25
Scottdale                               24
charlotte                               24
Newbury                                 23
Sharpsburg                              23
Robinson Twp.                           23
Cross Plains                            23
Mount Lebanon                           23
N Las Vegas                             22
Port Credit                             22
Rocky View                              22
Glendale Az                             22
Saint-JÃ©rÃ´me                          22
Saint-Jean-sur-Richelieu                22
Emsworth                                21
Lower Burrell                           21
Clairton                                21
Mont-Royal                              21
Stanley                                 21
Chambly                                 21
Montrose                                21
Cheswick                                20
Etna                                    20
Chestermere                             19
Harrison City                           19
Mc Murray                               19
Walton Hills                            19
Halton Hills                            19
Monongahela                             18
Midland                                 18
South Park                              18
Anjou                                   18
North Olmstead                          17
Blainville                              17
Greenfield Park                         17
Dollard-Des-Ormeaux                     16
Brooklin                                16
Ogden                                   16
Mt Lebanon                              16
Imperial                                16
St-BenoÃ®t de Mirabel                   15
LaSalle                                 15
Baldwin                                 15
St Joseph                               15
Ranlo                                   15
Montreal-Nord                           15
Export                                  15
LAS VEGAS                               15
Richmond Hil                            15
Don Mills                               14
Columbia Station                        14
Mc Farland                              14
Nellis AFB                              14
Kleinburg                               14
Brookpark                               14
McCandless Township                     14
New Eagle                               13
Olmsted Township                        13
Rouses Point                            13
Sainte-ThÃ©rÃ¨se                        13
Phx                                     13
Boisbriand                              13
Saint-Lazare                            12
Glassport                               12
Dollard-des Ormeaux                     12
Peters Township                         12
MontrÃ©al-Ouest                         12
Harrisbug                               12
Green Tree                              11
Saint-Eustache                          11
Elizabeth Township                      11
Sagamore Hills                          11
Mc Donald                               11
Indian land                             11
Lower burrell                           11
Presto                                  11
Jefferson Hills                         11
Sainte-Anne-De-Bellevue                 11
Terrebonne                              11
South Park Township                     11
Waddell                                 11
Avalon                                  11
Downtown                                11
Etibicoke                               10
Bainbridge                              10
Beaconsfield                            10
RosemÃ¨re                               10
Saint-Hyacinthe                         10
Sainte-AdÃ¨le                           10
East Liberty                            10
La Salle                                10
PHOENIX                                 10
Russellton                              10
Mont-Saint-Hilaire                      10
Lagrange                                10
Mint  Hill                              10
Lachute                                  9
Swissvale                                9
Medina Township                          9
Mentor On The Lake                       9
East Ajax                                9
Neville Island                           9
Paradise                                 9
Forest Hills                             9
North  York                              9
Warrensville Hts.                        9
Mantua                                   9
ChÃ¢teauguay                             8
Hinckley                                 8
N Ridgeville                             8
Schomberg                                8
Moseley                                  8
Balzac                                   8
Tuscola                                  8
Wilmerding                               8
Oakland                                  8
East Cleveland                           8
Bainbridge Township                      8
North Toronto                            7
MontrÃ©al-Nord                           7
Montreal-West                            7
Belle Vernon                             7
Nobleton                                 7
Central                                  7
Belleville                               7
Willowdale                               7
St-Jerome                                7
Mont-Saint-GrÃ©goire                     7
Moon TWP                                 7
Vimont                                   7
Rigaud                                   7
Mascouche                                7
Milton                                   7
Sheffield                                7
Library                                  7
Ville Mont-Royal                         7
Glen Williams                            7
Lowesville                               7
Ben Avon                                 6
Sunrise Manor                            6
Sheffield Lake                           6
solon                                    6
Brunswick Hills                          6
Cuddy                                    6
Goodwood                                 6
Shaler Township                          6
Sainte-Adele                             6
Whiitby                                  6
Venetia                                  6
Indianola                                6
Arnold                                   6
Saintt-Bruno-de-Montarville              6
Thornhil                                 5
Duquesne                                 5
Chateauguay                              5
Wilkins Township                         5
Dane                                     5
CommunautÃ©-Urbaine-de-MontrÃ©al         5
SCOTTSDALE                               5
GOODYEAR                                 5
Bellvue                                  5
Herminie                                 5
Mt Holly                                 5
Omaha                                    5
Kahnawake                                5
Bath                                     5
Sturgeon                                 5
Troy Township                            5
Sainte-Genevieve                         5
Saint-Leonard                            5
Mckees Rocks                             5
McCandless                               5
TORONTO                                  5
Mesa AZ                                  5
St-Leonard                               5
Mt. Holly                                5
Auburn Township                          4
Chargrin Falls                           4
Sainte-Julie                             4
NELLIS AFB                               4
Deforest                                 4
Saint-Marc-sur-Richelieu                 4
Ravenna                                  4
Dravosburg                               4
Mcfarland                                4
ETOBICOKE                                4
St Leonard                               4
King                                     4
Mcknight                                 4
Uxbridge                                 4
L'ile-Perrot                             4
Fort Mcdowell                            4
Saint-Bruno-de-Montarville               4
Beeton                                   4
Saint-Jerome                             4
Elrama                                   4
Litchfield                               4
Saint Joseph                             4
Dollard-Des Ormeaux                      4
St-Bruno-de-Montarville                  4
L'ÃŽle-Bizard                            4
Scarobrough                              4
Lawrence                                 4
Saint-Constant                           4
Lindale                                  4
Beloeil                                  4
Repentigny                               4
EstÃ©rel                                 4
Mcmurray                                 4
Homer                                    3
Delson                                   3
McDonald                                 3
Morin-Heights                            3
Central City                             3
South Amherst                            3
Painesville Township                     3
Canonsburd                               3
Mount Albert                             3
Paoli                                    3
Nellis Air Force Base                    3
Northwest Calgary                        3
Oakwood Village                          3
Central City Village                     3
Concord Township                         3
Rostraver                                3
Mirabel                                  3
Saint-Henri                              3
Fisher                                   3
Harwick                                  3
Mooers                                   3
Rural Ridge                              3
verdun                                   3
Champlain                                3
Lake Park                                3
River Drive Park                         3
Etobiicoke                               3
Montgomery                               3
East Mc Keesport                         3
Marshall                                 3
Gifford                                  2
Auburn Twp                               2
Northfield Center Township               2
Mount Oliver                             2
Mount Washington                         2
Midnapore                                2
Pointe-Calumet                           2
Creighton                                2
Mentor On the Lake                       2
Hyland Heights                           2
Saint Laurent                            2
Holland Landing                          2
Palgrave                                 2
kirtland                                 2
Saint-Basile-Le-Grand                    2
Schottsdale                              2
Pointe-Aux-Trembles                      2
Venise-en-QuÃ©bec                        2
Sainte-Marguerite-Esterel                2
Richmond Hts                             2
Thorncliffe Park                         2
Toronto Scarborough                      2
Rillton                                  2
Val-Morin                                2
Ross                                     2
Clark                                    2
Penn Hills Township                      2
East Pittsburgh                          2
Bois-des-Filion                          2
Joliette                                 2
Saint-Basile-le-Grand                    2
Cleveland Hghts.                         2
Hendersonville                           2
Mckeesport                               2
Sauk City                                2
Pincourt                                 2
Brookline                                2
Clarkson                                 2
Rocky View No. 44                        2
THORNHILL                                2
Waterloo                                 2
Candiac                                  2
Sainte-ThÃ©rÃ¨se-de-Blainville           2
Saint-Hippolyte                          2
Rosemere                                 2
Hemmingford                              2
MontrÃ©al-West                           1
North Braddock                           1
Oakwood                                  1
Hampstead                                1
Tottenham                                1
Saint-laurent                            1
Salaberry-De-Valleyfield                 1
Caledon Village                          1
Henryville                               1
Les Coteaux                              1
Cote Saint-Luc                           1
Port Vue                                 1
Sainte-Marthe                            1
Aliquippa                                1
Laval, Ste Dorothee                      1
Sainte-Catherine                         1
East McKeesport                          1
Sainte-Marguerite-du-lac-Masson          1
North Huntington                         1
Pitcairn                                 1
Cuyahoga Fls                             1
QuÃ©bec                                  1
Saint-Bruno                              1
Lawrenceville                            1
Salaberry-de-Valleyfield                 1
Caledon East                             1
Inglewood                                1
Fabreville                               1
Shorewood Hills                          1
Midway                                   1
Edgewood                                 1
cave creek                               1
Tolono                                   1
Malton                                   1
Beauharnois                              1
Ange-Gardien                             1
West Elizabeth                           1
Richmond Height                          1
Brownsburg-Chatham                       1
Iberville                                1
Montreal-Est                             1
Oak Ridges                               1
Rocky View County                        1
Kennedy Township                         1
Huntingdon                               1
Bratenahl                                1
Napierville                              1
Plum Boro                                1
De Winton                                1
St. LÃ©onard                             1
Mississuaga                              1
Buena Vista                              1
Rawdon                                   1
Black Earth                              1
North Strabane Township                  1
L'Assomption                             1
Ste-Rose                                 1
springdale                               1
Charlemagne                              1
Les CÃ¨dres                              1
Agincourt                                1
ÃŽle-des-Soeurs                          1
Mercier                                  1
Chatauguay                               1
Sun Praiie                               1
LaGrange                                 1
Coteau-du-Lac                            1
Deerfield                                1
Cuyahoga Heights                         1
Paw Creek                                1
(671,)

data = train_test_set.loc[train_test_set['city']!="Las Vegas", 'city']
weights = _np.ones(len(data)) / len(data)
_plt.figure(figsize=(20,10))
_plt.hist(data, weights=weights, bins=100)
_plt.title("City distribution")
_plt.gca().yaxis.set_major_formatter(_PercentFormatter(1))
_plt.show()

main_cities = city_count.where(city_count >= 100).dropna()
print(main_cities.to_string())
print(main_cities.shape)
main_cities = '|'.join(list(main_cities.index))

Las Vegas               208437.0
Phoenix                  71126.0
Toronto                  57047.0
Charlotte                40190.0
Scottsdale               36579.0
Pittsburgh               26891.0
Henderson                21518.0
MontrÃ©al                18531.0
Tempe                    18354.0
Mesa                     17235.0
Chandler                 15441.0
Gilbert                  14055.0
Cleveland                12873.0
Glendale                 10742.0
Madison                   9688.0
Markham                   8248.0
Calgary                   7958.0
Mississauga               7192.0
Peoria                    6582.0
North Las Vegas           4771.0
Surprise                  4117.0
Richmond Hill             3886.0
Goodyear                  3020.0
Concord                   2905.0
Scarborough               2544.0
Champaign                 2513.0
Vaughan                   2496.0
Avondale                  2451.0
Huntersville              2046.0
Lakewood                  2008.0
Matthews                  1986.0
North York                1977.0
Fort Mill                 1851.0
Brampton                  1643.0
Cave Creek                1538.0
Gastonia                  1282.0
Cornelius                 1278.0
Etobicoke                 1229.0
Akron                     1188.0
Westlake                  1118.0
Boulder City              1059.0
Thornhill                  985.0
Strongsville               957.0
Middleton                  930.0
Mentor                     864.0
Fountain Hills             836.0
Cuyahoga Falls             828.0
Rock Hill                  820.0
Cleveland Heights          813.0
Parma                      789.0
Kent                       745.0
Litchfield Park            745.0
Aurora                     712.0
North Olmsted              693.0
Beachwood                  654.0
Independence               648.0
Pineville                  647.0
Medina                     647.0
Newmarket                  637.0
Belmont                    631.0
Monroeville                610.0
Willoughby                 607.0
Rocky River                579.0
Solon                      576.0
Oakville                   573.0
Indian Trail               545.0
Avon                       543.0
Bridgeville                512.0
Davidson                   507.0
Hudson                     500.0
Pickering                  493.0
Laval                      491.0
Urbana                     491.0
Ajax                       479.0
Fitchburg                  464.0
Waxhaw                     462.0
Montreal                   441.0
Sun Prairie                434.0
Homestead                  429.0
Paradise Valley            425.0
Chagrin Falls              410.0
Stow                       406.0
Canonsburg                 400.0
Woodbridge                 374.0
Wexford                    368.0
Whitby                     367.0
Woodmere                   360.0
Carnegie                   358.0
Lyndhurst                  358.0
Coraopolis                 341.0
Fairlawn                   341.0
Bethel Park                331.0
Unionville                 326.0
Elyria                     319.0
Verona                     315.0
Berea                      313.0
Spring Valley              308.0
Sewickley                  306.0
Orange Village             297.0
Carefree                   294.0
Denver                     288.0
Brossard                   284.0
Brunswick                  282.0
Sun City                   280.0
Indian Land                267.0
Broadview Heights          259.0
Fairview Park              254.0
Monona                     253.0
Laveen                     253.0
Harrisburg                 252.0
Avon Lake                  249.0
Tolleson                   243.0
Twinsburg                  243.0
Mayfield Heights           242.0
Buckeye                    242.0
Middleburg Heights         239.0
Macedonia                  238.0
Dollard-des-Ormeaux        238.0
Brooklyn                   233.0
Oakmont                    224.0
Mint Hill                  224.0
Painesville                223.0
North Royalton             210.0
Kannapolis                 204.0
Amherst                    203.0
Queen Creek                203.0
Streetsboro                196.0
Monroe                     194.0
Clover                     193.0
El Mirage                  192.0
Orange                     188.0
Murrysville                188.0
Mount Holly                185.0
Verdun                     170.0
Dorval                     168.0
Maple                      167.0
Lake Wylie                 167.0
Chardon                    166.0
University Heights         165.0
Valley View                153.0
Highland Heights           153.0
South Las Vegas            153.0
Airdrie                    150.0
York                       149.0
Tega Cay                   148.0
Lorain                     147.0
West Mifflin               146.0
McMurray                   144.0
Moon Township              144.0
Chesterland                141.0
Euclid                     141.0
Allison Park               140.0
Irwin                      139.0
Brecksville                138.0
Westmount                  138.0
Waunakee                   137.0
Brook Park                 137.0
Stoughton                  134.0
Pointe-Claire              130.0
Northfield                 127.0
Warrensville Heights       127.0
Saint-Laurent              126.0
South Euclid               126.0
East York                  124.0
McKees Rocks               120.0
Bellevue                   117.0
Olmsted Falls              116.0
Braddock                   116.0
Blue Diamond               109.0
Ahwatukee                  109.0
New Kensington             108.0
Seven Hills                107.0
Mooresville                106.0
Upper Saint Clair          105.0
Longueuil                  105.0
North Ridgeville           102.0
(176,)

train_test_set['city'] = train_test_set['city'].str.findall(main_cities)
train_test_set['city'] = train_test_set['city'].map(lambda x: 'Other' if x==[] else x[0])
train_test_set.head()

We print and plot the distribution of the categories, to see the long tail and decide how many of them to keep.

category_count = _pd.Series(', '.join(list(train_test_set['categories'])).split(', ')).value_counts()
print(category_count.to_string())
print(category_count.shape)

Restaurants                         712379
Food                                192841
Nightlife                           170259
Bars                                165959
American (Traditional)              123975
Breakfast & Brunch                  120310
American (New)                      117437
Sandwiches                           82264
Mexican                              73726
Burgers                              71598
Pizza                                67538
Italian                              62490
Seafood                              57228
Japanese                             53952
Salad                                49388
Fast Food                            46982
Event Planning & Services            45382
Coffee & Tea                         43592
Chinese                              42971
Asian Fusion                         42576
Sushi Bars                           42487
Cafes                                40002
Steakhouses                          37097
Desserts                             36523
Cocktail Bars                        36152
Barbeque                             33004
Sports Bars                          26918
Vegetarian                           25616
Chicken Wings                        24833
Vegan                                23768
Wine Bars                            23685
Caterers                             23328
Mediterranean                        22625
Specialty Food                       21980
Thai                                 21702
Bakeries                             21321
Wine & Spirits                       21204
Beer                                 21204
Pubs                                 21076
Arts & Entertainment                 20724
Diners                               20452
Gastropubs                           17953
Korean                               17831
Gluten-Free                          17217
Buffets                              17133
Juice Bars & Smoothies               16826
Soup                                 16070
Noodles                              15555
Lounges                              15291
Delis                                14963
Vietnamese                           14470
Southern                             14139
Comfort Food                         13805
Tapas/Small Plates                   13335
French                               12767
Beer Bar                             12567
Venues & Event Spaces                12524
Canadian (New)                       12509
Indian                               12502
Tacos                                12369
Hotels & Travel                      11840
Middle Eastern                       11447
Ramen                                11015
Greek                                10833
Tex-Mex                              10615
Latin American                       10474
Hotels                               10261
Hawaiian                              9674
Ice Cream & Frozen Yogurt             9485
Breweries                             9293
Casinos                               9055
Tapas Bars                            8057
Ethnic Food                           7862
Chicken Shop                          6779
Hot Dogs                              6534
Cajun/Creole                          6237
Food Delivery Services                6237
Halal                                 5951
Poke                                  5903
Music Venues                          5663
Local Flavor                          5617
Soul Food                             5577
Shopping                              5421
Caribbean                             5311
Food Trucks                           5295
Bagels                                5287
Bubble Tea                            5119
Dim Sum                               5023
Grocery                               4705
Active Life                           4482
New Mexican Cuisine                   4385
Taiwanese                             4155
Creperies                             4008
Fish & Chips                          3903
Waffles                               3781
Resorts                               3552
Street Vendors                        3490
Pakistani                             3441
Spanish                               3393
Donuts                                3392
Tea Rooms                             3247
Wraps                                 3114
British                               3019
Dive Bars                             2989
Filipino                              2898
Cantonese                             2686
Dance Clubs                           2642
Karaoke                               2635
Brazilian                             2614
Cheesesteaks                          2594
Hot Pot                               2589
Irish                                 2566
Beauty & Spas                         2503
Modern European                       2479
Beer Gardens                          2269
Food Stands                           2191
Lebanese                              2170
Live/Raw Food                         2108
Smokehouse                            2062
Szechuan                              1985
Meat Shops                            1940
Arcades                               1925
African                               1904
Party & Event Planning                1852
German                                1814
Patisserie/Cake Shop                  1805
Health Markets                        1785
Poutineries                           1727
Day Spas                              1724
Imported Food                         1700
Izakaya                               1668
Brewpubs                              1638
Coffee Roasteries                     1622
Teppanyaki                            1577
Brasseries                            1461
Irish Pub                             1460
Peruvian                              1452
Pan Asian                             1441
Food Court                            1419
Internet Cafes                        1415
Persian/Iranian                       1402
Pasta Shops                           1357
Turkish                               1330
Acai Bowls                            1317
Kebab                                 1315
Falafel                               1275
Portuguese                            1260
Chocolatiers & Shops                  1258
Cuban                                 1257
Butcher                               1251
Bistros                               1237
Golf                                  1212
Seafood Markets                       1194
Whiskey Bars                          1173
Shopping Centers                      1146
Farmers Market                        1141
Automotive                            1130
Do-It-Yourself Food                   1122
Mongolian                             1073
Amusement Parks                       1066
Bowling                               1056
Conveyor Belt Sushi                   1053
Hookah Bars                           1051
Cupcakes                              1018
Fondue                                1005
Gelato                                1003
Polish                                1003
Singaporean                            965
Cinema                                 964
Shaved Ice                             951
Custom Cakes                           938
Malaysian                              926
Cheese Shops                           904
Afghan                                 900
Jazz & Blues                           884
Argentine                              870
Laotian                                848
Fruits & Veggies                       843
Performing Arts                        812
Kosher                                 808
Flowers & Gifts                        798
Organic Stores                         789
Colombian                              751
Venezuelan                             739
Convenience Stores                     735
Wedding Planning                       721
Arabian                                706
Japanese Curry                         705
Himalayan/Nepalese                     675
Belgian                                654
Ethiopian                              644
Fashion                                642
Car Wash                               641
Pool Halls                             588
Kids Activities                        584
Local Services                         582
Home & Garden                          569
Tuscan                                 551
Home Services                          542
Candy Stores                           522
Gas Stations                           501
Eatertainment                          489
Delicatessen                           475
Professional Services                  471
Russian                                468
Adult Entertainment                    467
Salvadoran                             467
Gift Shops                             459
Hair Salons                            453
International Grocery                  451
Health & Medical                       429
Moroccan                               426
Social Clubs                           407
Cafeteria                              381
Piano Bars                             376
Pretzels                               364
Hungarian                              360
Playgrounds                            352
Bed & Breakfast                        351
Ethnic Grocery                         349
Public Markets                         345
Speakeasies                            344
Beverage Store                         341
Art Galleries                          333
Hakka                                  326
Egyptian                               324
Pets                                   323
Puerto Rican                           310
Museums                                307
Ukrainian                              306
Macarons                               304
Education                              303
Wineries                               301
Empanadas                              300
Beer Garden                            291
Dinner Theater                         289
International                          288
Shaved Snow                            284
Nail Salons                            283
Arts & Crafts                          280
Indonesian                             265
Airports                               265
Personal Chefs                         263
Country Dance Halls                    263
Outlet Stores                          262
Musicians                              259
Public Services & Government           249
Hong Kong Style Cafe                   247
Real Estate                            245
Hostels                                245
Themed Cafes                           241
Syrian                                 236
Donairs                                236
Employment Agencies                    232
Furniture Stores                       231
Swimming Pools                         229
Massage                                228
Cambodian                              213
Department Stores                      213
Tours                                  211
Distilleries                           198
Dominican                              198
Mini Golf                              195
Zoos                                   195
Hair Removal                           191
Scandinavian                           183
Armenian                               179
Florists                               178
Shanghainese                           177
Festivals                              173
Waxing                                 172
Drugstores                             171
Kombucha                               171
Cooking Classes                        170
Tabletop Games                         170
Eyelash Service                        169
Personal Shopping                      168
Bangladeshi                            165
Nail Technicians                       165
Botanical Gardens                      165
Gay Bars                               163
Appliances                             160
Fitness & Instruction                  158
Towing                                 157
Barbers                                156
Wholesale Stores                       154
Sri Lankan                             154
Home Decor                             152
Landmarks & Historical Buildings       151
Nutritionists                          148
Kitchen & Bath                         146
Community Service/Non-Profit           145
Basque                                 137
Tiki Bars                              137
South African                          127
Wedding Chapels                        119
Books                                  117
Music & Video                          117
Mags                                   117
Antiques                               112
Sporting Goods                         110
Animal Shelters                        110
Skin Care                              110
Stadiums & Arenas                      109
Pop-Up Restaurants                     106
Guamanian                              106
Food Tours                             105
Tobacco Shops                          100
Cosmetics & Beauty Supply               99
Wine Tasting Room                       98
Laser Tag                               97
Herbs & Spices                          97
Printing Services                       94
Ethical Grocery                         93
Screen Printing                         93
Specialty Schools                       93
Walking Tours                           92
Recreation Centers                      91
Cabaret                                 89
Burmese                                 89
Cideries                                89
Udon                                    88
Coffee & Tea Supplies                   78
Gyms                                    78
Champagne Bars                          77
Street Art                              76
Austrian                                74
Baby Gear & Furniture                   74
Wigs                                    74
Health Retreats                         73
Australian                              73
Building Supplies                       72
Supper Clubs                            72
Doctors                                 72
Couriers & Delivery Services            67
Czech                                   67
Tasting Classes                         67
Auto Glass Services                     66
Country Clubs                           65
Yoga                                    65
Sports Wear                             64
Religious Organizations                 64
Horseback Riding                        61
Mass Media                              61
Shared Office Spaces                    61
Team Building Activities                60
Hobby Shops                             60
Shoe Stores                             60
Virtual Reality Centers                 60
Cooking Schools                         59
Veterinarians                           59
Pet Services                            59
Art Classes                             59
Haitian                                 59
Contractors                             56
Wholesalers                             56
Bartenders                              54
Outdoor Furniture Stores                53
Appliances & Repair                     52
Hospitals                               52
Accessories                             51
Parks                                   50
Uzbek                                   49
Financial Services                      49
Comedy Clubs                            48
Vehicle Wraps                           47
Farms                                   47
Honduran                                47
Art Museums                             47
Car Window Tinting                      47
Cultural Center                         47
Vacation Rentals                        47
Coffeeshops                             46
Women's Clothing                        45
Cigar Bars                              45
DJs                                     45
Vitamins & Supplements                  44
Dentists                                44
Trinidadian                             44
Trainers                                43
Bookstores                              43
Iberian                                 42
Emergency Medicine                      40
Skating Rinks                           40
Cardiologists                           40
Jewelry                                 40
Pediatricians                           40
Strip Clubs                             39
Auto Repair                             39
Sports Clubs                            39
Party Equipment Rentals                 39
Leisure Centers                         39
Indoor Playcentre                       39
Mattresses                              38
Hotel bar                               37
Colleges & Universities                 37
Nicaraguan                              37
Home Window Tinting                     37
Transportation                          37
Weight Loss Centers                     36
Party Supplies                          36
Pharmacy                                36
Bar Crawl                               36
Used                                    36
Vintage & Consignment                   36
Airport Terminals                       36
Men's Clothing                          34
Vinyl Records                           34
Cheese Tasting Classes                  34
Limos                                   34
Wine Tasting Classes                    33
Escape Games                            32
Rock Climbing                           32
Physical Therapy                        32
Dry Cleaning & Laundry                  32
Adult Education                         31
Cannabis Clinics                        31
RV Parks                                31
Real Estate Services                    31
Toy Stores                              30
General Dentistry                       30
Rehabilitation Center                   30
Sicilian                                30
CSA                                     29
Home Cleaning                           29
Pet Adoption                            28
Photography Stores & Services           28
Sugar Shacks                            28
Trampoline Parks                        28
Climbing                                28
Cards & Stationery                      27
Amateur Sports Teams                    25
Plumbing                                25
Popcorn Shops                           25
Cannabis Dispensaries                   25
Discount Store                          25
Hair Stylists                           25
Grilling Equipment                      24
Currency Exchange                       24
Cannabis Collective                     24
Check Cashing/Pay-day Loans             24
Fireplace Services                      23
Art Schools                             22
Opera & Ballet                          22
Security Systems                        22
Leather Goods                           22
Christmas Trees                         21
Car Rental                              21
Travel Services                         21
Post Offices                            20
Optometrists                            20
Auto Parts & Supplies                   20
Pet Stores                              20
Transmission Repair                     20
Bikes                                   20
Rotisserie Chicken                      19
Electronics                             19
Libraries                               19
Oil Change Stations                     19
Nurseries & Gardening                   19
Body Shops                              19
Auto Insurance                          19
Windshield Installation & Repair        19
Insurance                               19
Horse Racing                            19
Massage Therapy                         18
Videos & Video Game Rental              18
Family Practice                         17
Golf Lessons                            17
Makeup Artists                          17
Airport Shuttles                        16
Office Equipment                        16
Pool & Billiards                        16
Tennis                                  16
Bounce House Rentals                    16
Laundry Services                        16
Party Bus Rentals                       16
Eyebrow Services                        15
Dry Cleaning                            15
Security Services                       15
Race Tracks                             15
Personal Assistants                     15
Club Crawl                              15
Skydiving                               14
Sailing                                 14
Archery                                 14
Airport Lounges                         14
Laser Hair Removal                      14
Olive Oil                               14
Tanning                                 14
Video Game Stores                       13
Scottish                                13
Tanning Beds                            12
Pita                                    12
Interior Design                         12
Used Bookstore                          12
Spray Tanning                           12
Bingo Halls                             11
Yelp Events                             11
Reflexology                             11
Movers                                  11
Slovakian                               11
Boat Charters                           11
Town Car Service                        11
Tires                                   11
Reunion                                 11
Septic Services                         11
Lakes                                   11
Funeral Services & Cemeteries           11
Auto Customization                      11
Drive-Thru Bars                         11
Boating                                 11
Game Meat                               11
Ice Delivery                            11
Life Coach                              11
Summer Camps                            10
Soba                                    10
Golf Equipment                          10
Eyewear & Opticians                     10
Supernatural Readings                   10
Signmaking                              10
Psychics                                10
Mobile Phones                           10
Photographers                           10
Paint & Sip                             10
Ophthalmologists                        10
Hair Extensions                          9
Golf Cart Dealers                        9
Historical Tours                         9
Car Dealers                              9
Meditation Centers                       8
Oaxacan                                  8
Pick Your Own Farms                      8
Business Consulting                      8
Masonry/Concrete                         8
Reiki                                    8
Comic Books                              8
Psychologists                            8
Commercial Truck Repair                  8
Pumpkin Patches                          8
Bike Rentals                             8
Electricians                             8
Photo Booth Rentals                      8
Truck Rental                             8
Counseling & Mental Health               8
Graphic Design                           7
Mortgage Brokers                         7
Recording & Rehearsal Studios            7
Bus Tours                                7
Parenting Classes                        7
Guest Houses                             7
Tonkatsu                                 7
Preschools                               7
Bike Repair/Maintenance                  7
Head Shops                               7
Restaurant Supplies                      7
Ticket Sales                             7
Tempura                                  7
Dance Schools                            7
Real Estate Agents                       7
Audio/Visual Equipment Rental            6
Art Supplies                             6
Furniture Repair                         6
Pilates                                  6
Trophy Shops                             6
Souvenir Shops                           6
Piercing                                 6
Traditional Clothing                     6
Lighting Fixtures & Equipment            6
Bridal                                   6
Tattoo                                   6
Pool & Hot Tub Service                   6
Fur Clothing                             6
Beach Bars                               6
Chiropractors                            6
Martial Arts                             6
Furniture Rental                         6
Soccer                                   6
Carpet Installation                      5
Handyman                                 5
Medical Spas                             5
Pressure Washers                         5
Gutter Services                          5
Beaches                                  5
Acupuncture                              5
Hot Tub & Pool                           5
Tax Services                             5
Window Washing                           5
Windows Installation                     5
Heating & Air Conditioning/HVAC          5
Golf Equipment Shops                     5
Dog Walkers                              5
Pet Sitting                              5
Kids Hair Salons                         5
Property Management                      4
Door Sales/Installation                  4
Gun/Rifle Ranges                         4
Wine Tours                               4
Bankruptcy Law                           4
Bocce Ball                               4
Honey                                    4
Glass & Mirrors                          4
Lawyers                                  4
Hats                                     4
Knife Sharpening                         4
Software Development                     4
Tickets                                  4
Blow Dry/Out Services                    4
Keys & Locksmiths                        4
Water Stores                             4
Holistic Animal Care                     3
Calabrian                                3
Animal Assisted Therapy                  3
Animal Physical Therapy                  3
Swiss Food                               3
Aquarium Services                        3
Local Fish Stores                        3
Diagnostic Services                      3
Brazilian Jiu-jitsu                      3
Mauritius                                3
Laboratory Testing                       3
Thrift Stores                            3
Cosmetic Dentists                        3
Surf Schools                             3
Tai Chi                                  3
Vape Shops                               3
Private Tutors                           3
Observatories                            3
Ski Resorts                              3
Web Design                               3
Aquariums                                3
Studio Taping                            3
Churches                                 3
Advertising                              3
Food Banks                               3
Hardware Stores                          3
Day Camps                                3
Apartments                               3
Occupational Therapy                     3
Airsoft                                  3
Pet Training                             3
Swimwear                                 3
Hainan                                   3
Accountants                              3
Unofficial Yelp Events                   3
Police Departments                       3
Pop-up Shops                             3
Air Duct Cleaning                        3
Paint-Your-Own Pottery                   3
Aircraft Repairs                         3
Flea Markets                             2
Pub Food                                 2
Divorce & Family Law                     2
Engraving                                2
Naturopathic/Holistic                    2
Marinas                                  2
Plus Size Fashion                        2
Immigration Law                          2
Computers                                2
Wills                                    2
Magicians                                2
Medical Cannabis Referrals               2
Car Share Services                       2
Bespoke Clothing                         2
Public Transportation                    2
Hunting & Fishing Supplies               2
Boat Dealers                             2
Medical Centers                          2
Trusts                                   2
Tax Law                                  2
Community Centers                        2
Boat Repair                              2
Outdoor Gear                             2
Pet Boarding                             2
Formal Wear                              2
Marketing                                2
Estate Planning Law                      2
Painters                                 2
Courthouses                              2
& Probates                               2
Service Stations                         2
Personal Injury Law                      2
Drywall Installation & Repair            2
Clowns                                   2
Batting Cages                            1
Beer Hall                                1
Holiday Decorations                      1
Furniture Reupholstery                   1
Churros                                  1
Clothing Rental                          1
University Housing                       1
Pet Groomers                             1
Auto Detailing                           1
Pawn Shops                               1
Senegalese                               1
Gardeners                                1
Campgrounds                              1
Banks & Credit Unions                    1
Laundromat                               1
RV Repair                                1
Roofing                                  1
Officiants                               1
Music & DVDs                             1
Home Health Care                         1
Squash                                   1
Auto Upholstery                          1
Siding                                   1
Landscaping                              1
Motorcycle Repair                        1
Ski Schools                              1
Flooring                                 1
Taxis                                    1
Foundation Repair                        1
Minho                                    1
(714,)

data = category_count.drop(labels=['Restaurants', 'Food']).index
vals = category_count.drop(labels=['Restaurants', 'Food']).values
weights = vals / vals.sum()
_plt.figure(figsize=(20,10))
_plt.hist(data, weights=weights, bins=100)
_plt.title("Category distribution")
_plt.gca().yaxis.set_major_formatter(_PercentFormatter(1))
_plt.show()

main_categories = category_count.drop(labels=['Restaurants', 'Food']).where(category_count >= 100).dropna()
print(main_categories.to_string())
print(main_categories.shape)
main_categories = '|'.join([_re.escape(x) for x in main_categories.index])

Nightlife                           170259.0
Bars                                165959.0
American (Traditional)              123975.0
Breakfast & Brunch                  120310.0
American (New)                      117437.0
Sandwiches                           82264.0
Mexican                              73726.0
Burgers                              71598.0
Pizza                                67538.0
Italian                              62490.0
Seafood                              57228.0
Japanese                             53952.0
Salad                                49388.0
Fast Food                            46982.0
Event Planning & Services            45382.0
Coffee & Tea                         43592.0
Chinese                              42971.0
Asian Fusion                         42576.0
Sushi Bars                           42487.0
Cafes                                40002.0
Steakhouses                          37097.0
Desserts                             36523.0
Cocktail Bars                        36152.0
Barbeque                             33004.0
Sports Bars                          26918.0
Vegetarian                           25616.0
Chicken Wings                        24833.0
Vegan                                23768.0
Wine Bars                            23685.0
Caterers                             23328.0
Mediterranean                        22625.0
Specialty Food                       21980.0
Thai                                 21702.0
Bakeries                             21321.0
Wine & Spirits                       21204.0
Beer                                 21204.0
Pubs                                 21076.0
Arts & Entertainment                 20724.0
Diners                               20452.0
Gastropubs                           17953.0
Korean                               17831.0
Gluten-Free                          17217.0
Buffets                              17133.0
Juice Bars & Smoothies               16826.0
Soup                                 16070.0
Noodles                              15555.0
Lounges                              15291.0
Delis                                14963.0
Vietnamese                           14470.0
Southern                             14139.0
Comfort Food                         13805.0
Tapas/Small Plates                   13335.0
French                               12767.0
Beer Bar                             12567.0
Venues & Event Spaces                12524.0
Canadian (New)                       12509.0
Indian                               12502.0
Tacos                                12369.0
Hotels & Travel                      11840.0
Middle Eastern                       11447.0
Ramen                                11015.0
Greek                                10833.0
Tex-Mex                              10615.0
Latin American                       10474.0
Hotels                               10261.0
Hawaiian                              9674.0
Ice Cream & Frozen Yogurt             9485.0
Breweries                             9293.0
Casinos                               9055.0
Tapas Bars                            8057.0
Ethnic Food                           7862.0
Chicken Shop                          6779.0
Hot Dogs                              6534.0
Cajun/Creole                          6237.0
Food Delivery Services                6237.0
Halal                                 5951.0
Poke                                  5903.0
Music Venues                          5663.0
Local Flavor                          5617.0
Soul Food                             5577.0
Shopping                              5421.0
Caribbean                             5311.0
Food Trucks                           5295.0
Bagels                                5287.0
Bubble Tea                            5119.0
Dim Sum                               5023.0
Grocery                               4705.0
Active Life                           4482.0
New Mexican Cuisine                   4385.0
Taiwanese                             4155.0
Creperies                             4008.0
Fish & Chips                          3903.0
Waffles                               3781.0
Resorts                               3552.0
Street Vendors                        3490.0
Pakistani                             3441.0
Spanish                               3393.0
Donuts                                3392.0
Tea Rooms                             3247.0
Wraps                                 3114.0
British                               3019.0
Dive Bars                             2989.0
Filipino                              2898.0
Cantonese                             2686.0
Dance Clubs                           2642.0
Karaoke                               2635.0
Brazilian                             2614.0
Cheesesteaks                          2594.0
Hot Pot                               2589.0
Irish                                 2566.0
Beauty & Spas                         2503.0
Modern European                       2479.0
Beer Gardens                          2269.0
Food Stands                           2191.0
Lebanese                              2170.0
Live/Raw Food                         2108.0
Smokehouse                            2062.0
Szechuan                              1985.0
Meat Shops                            1940.0
Arcades                               1925.0
African                               1904.0
Party & Event Planning                1852.0
German                                1814.0
Patisserie/Cake Shop                  1805.0
Health Markets                        1785.0
Poutineries                           1727.0
Day Spas                              1724.0
Imported Food                         1700.0
Izakaya                               1668.0
Brewpubs                              1638.0
Coffee Roasteries                     1622.0
Teppanyaki                            1577.0
Brasseries                            1461.0
Irish Pub                             1460.0
Peruvian                              1452.0
Pan Asian                             1441.0
Food Court                            1419.0
Internet Cafes                        1415.0
Persian/Iranian                       1402.0
Pasta Shops                           1357.0
Turkish                               1330.0
Acai Bowls                            1317.0
Kebab                                 1315.0
Falafel                               1275.0
Portuguese                            1260.0
Chocolatiers & Shops                  1258.0
Cuban                                 1257.0
Butcher                               1251.0
Bistros                               1237.0
Golf                                  1212.0
Seafood Markets                       1194.0
Whiskey Bars                          1173.0
Shopping Centers                      1146.0
Farmers Market                        1141.0
Automotive                            1130.0
Do-It-Yourself Food                   1122.0
Mongolian                             1073.0
Amusement Parks                       1066.0
Bowling                               1056.0
Conveyor Belt Sushi                   1053.0
Hookah Bars                           1051.0
Cupcakes                              1018.0
Fondue                                1005.0
Gelato                                1003.0
Polish                                1003.0
Singaporean                            965.0
Cinema                                 964.0
Shaved Ice                             951.0
Custom Cakes                           938.0
Malaysian                              926.0
Cheese Shops                           904.0
Afghan                                 900.0
Jazz & Blues                           884.0
Argentine                              870.0
Laotian                                848.0
Fruits & Veggies                       843.0
Performing Arts                        812.0
Kosher                                 808.0
Flowers & Gifts                        798.0
Organic Stores                         789.0
Colombian                              751.0
Venezuelan                             739.0
Convenience Stores                     735.0
Wedding Planning                       721.0
Arabian                                706.0
Japanese Curry                         705.0
Himalayan/Nepalese                     675.0
Belgian                                654.0
Ethiopian                              644.0
Fashion                                642.0
Car Wash                               641.0
Pool Halls                             588.0
Kids Activities                        584.0
Local Services                         582.0
Home & Garden                          569.0
Tuscan                                 551.0
Home Services                          542.0
Candy Stores                           522.0
Gas Stations                           501.0
Eatertainment                          489.0
Delicatessen                           475.0
Professional Services                  471.0
Russian                                468.0
Adult Entertainment                    467.0
Salvadoran                             467.0
Gift Shops                             459.0
Hair Salons                            453.0
International Grocery                  451.0
Health & Medical                       429.0
Moroccan                               426.0
Social Clubs                           407.0
Cafeteria                              381.0
Piano Bars                             376.0
Pretzels                               364.0
Hungarian                              360.0
Playgrounds                            352.0
Bed & Breakfast                        351.0
Ethnic Grocery                         349.0
Public Markets                         345.0
Speakeasies                            344.0
Beverage Store                         341.0
Art Galleries                          333.0
Hakka                                  326.0
Egyptian                               324.0
Pets                                   323.0
Puerto Rican                           310.0
Museums                                307.0
Ukrainian                              306.0
Macarons                               304.0
Education                              303.0
Wineries                               301.0
Empanadas                              300.0
Beer Garden                            291.0
Dinner Theater                         289.0
International                          288.0
Shaved Snow                            284.0
Nail Salons                            283.0
Arts & Crafts                          280.0
Indonesian                             265.0
Airports                               265.0
Personal Chefs                         263.0
Country Dance Halls                    263.0
Outlet Stores                          262.0
Musicians                              259.0
Public Services & Government           249.0
Hong Kong Style Cafe                   247.0
Real Estate                            245.0
Hostels                                245.0
Themed Cafes                           241.0
Syrian                                 236.0
Donairs                                236.0
Employment Agencies                    232.0
Furniture Stores                       231.0
Swimming Pools                         229.0
Massage                                228.0
Cambodian                              213.0
Department Stores                      213.0
Tours                                  211.0
Distilleries                           198.0
Dominican                              198.0
Mini Golf                              195.0
Zoos                                   195.0
Hair Removal                           191.0
Scandinavian                           183.0
Armenian                               179.0
Florists                               178.0
Shanghainese                           177.0
Festivals                              173.0
Waxing                                 172.0
Drugstores                             171.0
Kombucha                               171.0
Cooking Classes                        170.0
Tabletop Games                         170.0
Eyelash Service                        169.0
Personal Shopping                      168.0
Bangladeshi                            165.0
Nail Technicians                       165.0
Botanical Gardens                      165.0
Gay Bars                               163.0
Appliances                             160.0
Fitness & Instruction                  158.0
Towing                                 157.0
Barbers                                156.0
Wholesale Stores                       154.0
Sri Lankan                             154.0
Home Decor                             152.0
Landmarks & Historical Buildings       151.0
Nutritionists                          148.0
Kitchen & Bath                         146.0
Community Service/Non-Profit           145.0
Basque                                 137.0
Tiki Bars                              137.0
South African                          127.0
Wedding Chapels                        119.0
Books                                  117.0
Music & Video                          117.0
Mags                                   117.0
Antiques                               112.0
Sporting Goods                         110.0
Animal Shelters                        110.0
Skin Care                              110.0
Stadiums & Arenas                      109.0
Pop-Up Restaurants                     106.0
Guamanian                              106.0
Food Tours                             105.0
Tobacco Shops                          100.0
(306,)

train_test_set['categories'] = train_test_set['categories'].str.findall(main_categories)
train_test_set['categories'] = train_test_set['categories'].map(lambda x: set(x))
train_test_set['categories'] = train_test_set['categories'].map(lambda x: ['Other'] if not bool(x) else list(x))
train_test_set['categories'] = train_test_set['categories'].map(', '.join) 
train_test_set.head()

Now we apply the actual conversion

train_test_set.shape

(712379, 93)

cat_cols = ['OutdoorSeating', 'BusinessAcceptsCreditCards', 'RestaurantsDelivery', 'RestaurantsReservations', 'WiFi',
        'Alcohol', 'city']
train_test_set = _pd.get_dummies(train_test_set, columns=cat_cols, prefix=cat_cols)
train_test_set.shape

(712379, 281)

categories = train_test_set['categories'].str.get_dummies(',')
f1 = lambda x: "categories_" + x
categories.rename(columns=f1, inplace=True)
train_test_set[categories.columns] = categories
train_test_set.drop(columns=['categories'], inplace=True)
train_test_set.shape

(712379, 787)

oe = _OrdinalEncoder()

ord_cols = ['Monday_Open', 'Tuesday_Open', 'Wednesday_Open', 'Thursday_Open', 'Friday_Open',
            'Saturday_Open', 'Sunday_Open', 'Monday_Close', 'Tuesday_Close', 'Wednesday_Close',
            'Thursday_Close','Friday_Close', 'Saturday_Close', 'Sunday_Close', 'postal_code']

train_test_set[ord_cols] = oe.fit_transform(train_test_set[ord_cols].to_numpy())

The resulting dataset

train_test_set.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 712379 entries, 0 to 153992
Columns: 787 entries, review_id to categories_Wraps
dtypes: float64(68), int32(1), int64(520), object(3), uint8(195)
memory usage: 3.3+ GB

train_set = train_test_set[:_train_len]
test_set = train_test_set[_train_len:]

train_set.head(10)

train_set.shape

(558386, 787)

train_set.to_pickle('../dataset/m2_n9/model_train_set_3.pickle')

test_set.head(10)

test_set.shape

(153993, 787)

test_set.to_pickle('../dataset/m2_n9/model_test_set_3.pickle')

_del_all()

6. Models¶

6.1 Linear SVM¶

(see the docs)

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')
train_set.head()

#sub_train_set = train_set[:round(train_set.shape[0]/3)]
sub_train_set = train_set
del train_set
sub_train_set.shape

(558386, 817)

# define classifier
svc_classifier = _LinearSVC(random_state=0, max_iter=50000)
svc_classifier.get_params()

{'C': 1.0,
 'class_weight': None,
 'dual': True,
 'fit_intercept': True,
 'intercept_scaling': 1,
 'loss': 'squared_hinge',
 'max_iter': 50000,
 'multi_class': 'ovr',
 'penalty': 'l2',
 'random_state': 0,
 'tol': 0.0001,
 'verbose': 0}

# fine tune classifier
# param_grid = {'C':[0.001,0.01,0.1,0.25,0.5,0.75,1,10,100,1000], 'gamma':[3,2,1,0.1,0.001,0.0001]}
param_grid = {'C':[0.001,0.01,0.1,0.25,0.5,0.75,1,10,100,1000]}
# grid = _GridSearchCV(estimator=svc_classifier, param_grid=param_grid, refit=True, verbose=2, cv=3, error_score=_np.nan, n_jobs=1, pre_dispatch=1)
grid = _GridSearchCV(estimator=svc_classifier, param_grid=param_grid, refit=True, verbose=2, cv=3, error_score=_np.nan, n_jobs=-1, pre_dispatch=6)
grid.fit(sub_train_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), sub_train_set['likes'])
print("best params:", grid.best_params_, "- best score:", grid.best_score_)

Fitting 3 folds for each of 10 candidates, totalling 30 fits

[Parallel(n_jobs=-1)]: Using backend LokyBackend with 12 concurrent workers.
[Parallel(n_jobs=-1)]: Done  30 out of  30 | elapsed: 1654.8min finished

best params: {'C': 0.001} - best score: 0.7436808945783024

C:\Users\super\Anaconda3\envs\tf-gpu\lib\site-packages\sklearn\svm\base.py:929: ConvergenceWarning: Liblinear failed to converge, increase the number of iterations.
  "the number of iterations.", ConvergenceWarning)

print("results:", grid.cv_results_)

results: {'mean_fit_time': array([19900.99567477, 19961.755548  , 19843.87850094, 19883.72761846,
       19875.85472822, 19892.886506  , 19819.18020558, 19831.93277272,
       19672.39210947, 19633.8438464 ]), 'std_fit_time': array([11.92322164, 24.47909948, 11.8002273 ,  2.51810785, 18.78170298,
       16.00907826,  5.86167115, 32.23873155,  4.59971461, 24.49313577]), 'mean_score_time': array([ 2.63927579,  5.25461706,  5.47336515, 10.05810801,  5.17248909,
        3.84837699,  2.75297205,  2.87099091,  2.33608754,  1.75242146]), 'std_score_time': array([0.28703641, 3.56587212, 1.62402285, 4.61803914, 2.93643879,
       1.57086447, 0.04191375, 0.27929337, 0.24098951, 0.2409035 ]), 'param_C': masked_array(data=[0.001, 0.01, 0.1, 0.25, 0.5, 0.75, 1, 10, 100, 1000],
             mask=[False, False, False, False, False, False, False, False,
                   False, False],
       fill_value='?',
            dtype=object), 'params': [{'C': 0.001}, {'C': 0.01}, {'C': 0.1}, {'C': 0.25}, {'C': 0.5}, {'C': 0.75}, {'C': 1}, {'C': 10}, {'C': 100}, {'C': 1000}], 'split0_test_score': array([0.74414656, 0.68984581, 0.71937893, 0.57199269, 0.71992156,
       0.70982109, 0.59223124, 0.63179498, 0.63454575, 0.63445441]), 'split1_test_score': array([0.74462735, 0.7388034 , 0.70227478, 0.59544507, 0.49747486,
       0.62558025, 0.65421645, 0.68013947, 0.66420958, 0.66396781]), 'split2_test_score': array([0.74226876, 0.72382984, 0.55883048, 0.61401831, 0.47335167,
       0.52100705, 0.69141129, 0.52799149, 0.52916273, 0.5987976 ]), 'mean_test_score': array([0.74368089, 0.71749292, 0.66016161, 0.59381861, 0.56358326,
       0.61880312, 0.6459528 , 0.61330871, 0.60930611, 0.63240661]), 'std_test_score': array([0.00101764, 0.020483  , 0.07199118, 0.0171954 , 0.11098627,
       0.0772319 , 0.04090958, 0.06347462, 0.0579493 , 0.02664495]), 'rank_test_score': array([ 1,  2,  3,  9, 10,  6,  4,  7,  8,  5])}

del sub_train_set
train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')

best_model = grid.best_estimator_
best_model.set_params(verbose=5)
best_model.get_params()

{'C': 0.001,
 'class_weight': None,
 'dual': True,
 'fit_intercept': True,
 'intercept_scaling': 1,
 'loss': 'squared_hinge',
 'max_iter': 50000,
 'multi_class': 'ovr',
 'penalty': 'l2',
 'random_state': 0,
 'tol': 0.0001,
 'verbose': 5}

best_model.fit(train_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])

C:\Users\super\Anaconda3\envs\tf-gpu\lib\site-packages\sklearn\svm\base.py:929: ConvergenceWarning: Liblinear failed to converge, increase the number of iterations.
  "the number of iterations.", ConvergenceWarning)

LinearSVC(C=0.001, class_weight=None, dual=True, fit_intercept=True,
          intercept_scaling=1, loss='squared_hinge', max_iter=50000,
          multi_class='ovr', penalty='l2', random_state=0, tol=0.0001,
          verbose=0)

print("coef:", best_model.coef_)
print("intercept:", best_model.intercept_)

coef: [[-1.43167078e-01 -2.03906910e-01  3.42779107e-01 -1.04728908e-02
  -1.47427540e-01 -2.43472370e-02 -9.26808435e-03 -1.17709493e-02
   5.58234365e-02  1.83726109e-02 -7.91658762e-03  7.49998933e-05
  -3.29092790e-02  7.08513173e-03 -4.64203658e-05  6.24335229e-01
  -4.68312019e-05  1.13214836e-03  1.87475912e-04 -8.54227633e-04
  -4.47446153e-04 -7.61677236e-04  1.13923954e-03 -1.91138083e-04
  -1.40924026e-04  4.50836058e-04 -7.61383085e-05  5.97650915e-05
   2.98315416e-05 -2.74831993e-04 -6.26183842e-02  5.30331089e-05
   5.84330552e-03  2.03076934e-04 -3.48781537e-02 -2.18059846e-04
  -4.02326827e-02  4.95315660e-02  3.39012708e-04  1.71206420e-02
  -8.71542990e-04 -2.61593855e-04  1.30803827e-04  1.18480402e-04
  -4.80787054e-06 -5.59617233e-03  1.61222017e-02  1.03926925e-04
   4.67182627e-03  5.79030033e-03  3.74660499e-02  1.02789779e-02
   4.51314090e-02  1.57159326e-02  2.23464027e-02  5.36922329e-02
   2.34749232e-02  2.86587882e-02  1.87929188e-02  3.83956193e-04
  -7.22989332e-03  1.95846857e-03  1.82299621e-03  1.13746379e-03
   5.45619480e-03  1.58677886e-02 -1.95511703e-02 -1.28162672e-02
  -1.96649484e-03 -1.30449955e-03 -8.07076837e-04  1.18903675e-02
  -1.11566651e-02  2.26932299e-02  2.51582368e-02 -1.87668524e-02
   7.85823969e-03  7.73799384e-03  3.85827228e-03  4.22093591e-02
  -1.19005102e-01 -8.59189500e-02 -1.24239801e-01 -7.40342751e-02
  -1.19770059e-01 -1.35359519e-01 -1.20415320e-01 -1.05257346e-01
  -1.03491187e-01 -1.14153703e-01 -1.08812496e-01 -1.06197655e-01
  -8.22293134e-02 -8.11447278e-02 -8.84894517e-02 -7.73003600e-02
  -8.49805052e-02 -8.51293957e-02 -8.21027373e-02 -7.69512147e-02
   6.43923766e-03  1.85595762e-02  1.23813296e-02  2.25462539e-02
   2.14381375e-02 -2.21330989e-02  1.39767176e-02  3.28450365e-02
  -5.99616006e-02  4.27423193e-02  1.95000974e-02 -7.25494945e-02
  -1.51045288e-02  3.20011680e-02  2.51968573e-02  6.58713427e-02
  -9.68960933e-03 -7.67558814e-02 -1.31860158e-02 -3.30225057e-02
  -1.29565963e-03 -1.22750226e-03  1.32295606e-02  2.79374411e-02
   3.17061198e-03  4.61489025e-03  3.31614833e-01 -6.92978644e-03
   1.75307888e-02 -8.55135160e-03  2.19597610e-02  1.50690523e-02
   4.00123219e-02 -1.79436834e-02  2.37032470e-02 -1.74807415e-01
   2.08777300e-02  5.59448146e-02 -2.29348176e-02 -1.07672262e-01
   1.36945435e-03 -7.92229306e-02  2.30166566e-02 -2.07064407e-02
  -5.24655802e-02  1.29004963e-02  6.08488450e-03  1.29702210e-02
  -1.07647542e-02  1.95574537e-02 -6.24734869e-02  1.80610497e-02
   1.38307854e-02  7.02867332e-03  3.03232666e-02 -1.18367338e-01
   6.66244523e-03 -1.01049018e-01 -2.26197994e-02  2.09783558e-03
  -9.12480404e-03 -1.16218404e-02  6.37714957e-02  2.12609828e-02
   2.93436435e-02  1.84281145e-02 -8.92567984e-02  2.20758890e-02
  -3.20071304e-02 -7.40689432e-02  1.19849983e-02 -4.26474567e-02
   3.20709228e-02 -1.48892296e-02  5.93652875e-02  8.70463297e-02
   1.68511943e-02  1.08955283e-02  1.15257076e-02 -1.13721061e-02
   4.61982521e-03 -5.38403883e-03  1.59303884e-02  1.21825364e-01
  -1.43542730e-02 -1.70066160e-02 -1.28769520e-01 -1.08798946e-02
  -2.89941634e-02  6.84521535e-03  4.20171560e-02  6.75656292e-02
  -2.54647582e-02  2.03317828e-02  1.13335713e-01 -2.34771344e-02
  -3.49674158e-02  2.87547751e-02 -1.93415639e-02 -3.62824126e-02
  -1.92791682e-03  5.32535767e-02 -1.42032229e-02 -2.13908527e-02
  -8.23161692e-02  1.29157539e-02 -2.13631810e-02 -1.48627573e-02
   7.81676486e-02 -1.76906490e-02  2.07163109e-02  5.49162626e-03
  -8.29327282e-02  3.36895336e-02 -1.16082899e-02 -5.93015570e-02
   1.42680472e-02 -8.77579215e-03  1.86299090e-03 -3.94405655e-03
  -5.30170043e-03 -2.23817348e-02  6.64656794e-02 -1.46158840e-02
  -7.95928591e-03  1.54178914e-02 -1.08412578e-01 -1.14603015e-02
   1.15143000e-02  3.07233178e-03 -1.39648332e-02 -8.49316438e-02
   9.31174193e-03  1.82254249e-02 -4.93272251e-02 -1.52058544e-02
  -3.68582782e-04 -2.93260038e-02  4.72522780e-02 -1.72815316e-02
   4.17304527e-02  8.30807812e-03 -6.71044760e-03  3.67725198e-02
   3.40228374e-02 -9.29651532e-03 -1.14372482e-02  4.50867347e-02
   1.25293732e-02 -5.20908397e-02 -3.11803374e-02 -3.66637724e-02
  -3.40471406e-02 -8.13300220e-02  5.83104557e-03 -6.15754552e-03
  -3.83566701e-03 -3.85224902e-03  2.56349472e-02  3.97811814e-03
  -4.23112661e-02 -1.22968513e-02  6.85746928e-02  1.49594730e-02
   2.42777204e-02 -4.57823769e-02 -4.68553937e-02  8.88235693e-03
  -2.69917143e-02  6.05564616e-03  3.70784601e-03  6.36277093e-03
  -1.96343197e-02  2.50612973e-02 -8.80448869e-03 -4.39302037e-02
  -2.70567179e-02  1.28431973e-03  1.82585550e-02 -2.99224173e-02
  -3.76426445e-02 -1.21557100e-02  6.10713723e-04  6.84306098e-03
   6.13247044e-03  1.48782541e-02  1.67792883e-02  5.42475598e-03
   1.67573925e-02 -7.66087771e-04 -1.00430850e-02 -2.13891203e-02
  -8.00217272e-03 -1.27645522e-02 -2.72593723e-03  5.15850376e-03
  -5.24853616e-02  2.93798828e-03  1.16753929e-02 -6.64935227e-05
  -1.01851634e-02 -2.53813056e-02  4.83795523e-04  3.51256411e-02
  -9.64920258e-03  3.62038039e-03 -6.03079323e-02  6.03129133e-03
   1.82823595e-03  4.00928974e-03  1.93399507e-02  1.64871125e-02
  -9.22545284e-04  1.45307420e-02  2.93889782e-03  1.46391268e-03
   4.21477946e-03  1.95149473e-02  1.59151458e-02 -2.90799407e-02
  -5.32425108e-03  5.13371472e-04  6.58654665e-03  2.21270704e-02
  -2.92310377e-03  1.87968959e-02  6.41981957e-02 -5.36022064e-02
   1.64240343e-02 -7.36316514e-03  1.74012276e-02 -3.43569952e-02
  -3.33796058e-02 -8.25664541e-03  7.68784960e-03  1.79893762e-02
  -1.23231879e-02  3.42456239e-03 -8.12921342e-02  1.18766670e-03
   7.81862807e-03 -1.61176409e-03 -7.26807030e-03  5.20989335e-02
  -2.21053123e-02 -1.06174103e-02  2.09679791e-03 -2.27182601e-02
   1.28459985e-03 -1.60854286e-02  4.08880864e-02  8.12019316e-03
   4.77563249e-02 -2.39699682e-02  1.54578864e-02 -8.06455222e-03
  -3.49687057e-02  7.48646031e-03 -1.01357608e-02  1.46504891e-02
   8.00686052e-03  8.40864088e-03 -6.75660634e-03  5.74274899e-03
   6.26943831e-02  5.51697698e-03 -8.23761484e-04 -1.00452237e-03
   7.99939968e-03 -2.46815672e-02  2.66978365e-02  2.21327073e-02
  -8.80816123e-03  7.18237663e-03  6.57952294e-03  5.02076932e-02
  -3.91555575e-03 -5.57597968e-03  1.46251006e-02  5.72630443e-03
   1.16652072e-02 -1.53468443e-02 -1.92508581e-03 -5.60256986e-02
  -1.12202980e-02  6.80762675e-02  4.41100854e-02 -7.29487438e-03
  -1.31907835e-03 -4.49189138e-02  6.87885210e-03  9.98582527e-03
  -2.25761715e-02  3.39400124e-03  2.20140659e-03 -1.60394518e-02
   2.53522844e-02  2.66762932e-03 -5.18357602e-03  1.56388295e-02
  -1.79071588e-03  2.77794683e-02 -3.67316202e-02 -1.42449595e-03
  -6.40470001e-02 -1.55041448e-02  1.13679417e-02 -2.17145211e-02
  -7.53775575e-03  1.22737060e-02  8.65845652e-04 -3.08407608e-03
   4.86664233e-03 -1.53867109e-02 -5.17985113e-03 -2.03651799e-02
   1.91980806e-02  4.92713278e-03 -6.53381963e-03 -1.81574415e-02
   9.65854970e-05 -3.01428875e-02  1.26714661e-02 -1.76560612e-02
   6.68903647e-03 -9.93572804e-03 -3.58821993e-02  8.32218218e-03
  -4.36526108e-03  2.48863141e-02  3.61363130e-02  5.07821855e-03
  -3.90056078e-03  9.36203437e-03 -3.29313219e-02 -3.11253544e-02
   1.15892173e-02 -5.27904735e-02 -1.62300116e-03  2.91537406e-02
  -3.29570759e-02  3.94202889e-03 -2.01852252e-02 -4.79670044e-03
   7.28264419e-04  7.18623583e-03  7.90700878e-03  2.44060085e-02
   1.00612486e-03 -2.23966525e-02 -3.79289197e-03 -1.10582143e-02
   1.46512937e-02  2.44656784e-02 -1.15407013e-02 -5.53300392e-03
   2.42208289e-03 -4.41613655e-02  6.48285154e-03  4.13872823e-02
   9.99939903e-03 -1.18352048e-02 -2.40933106e-02  2.26273085e-02
  -1.33455480e-03  3.15784295e-02 -2.48895846e-03  1.22946390e-02
   2.42208289e-03 -1.44142302e-02 -8.74927111e-04 -3.20937165e-03
   7.39171882e-03 -1.02028844e-02  2.26375269e-02  3.31014263e-03
  -8.40897442e-04  2.45346628e-02 -6.63887082e-02 -2.90423744e-03
   3.04557214e-02 -1.90037208e-02 -2.61113659e-02  3.80622415e-02
   7.87264893e-03  3.26674265e-03  2.17321140e-02  1.02955744e-02
   2.83061855e-03  2.63610086e-02  7.77730787e-02  1.05384436e-02
   3.17608021e-03 -2.73779197e-03  5.11149650e-03  2.87392776e-03
  -1.01253861e-03 -1.08366221e-03 -1.30241635e-03  1.20216849e-02
   1.39569501e-02 -1.99490256e-02  1.35320484e-02  8.05746616e-03
  -8.24048119e-03  2.61876051e-02  5.26147434e-03 -1.21678933e-03
   1.17283157e-02 -1.29200471e-02  1.47292929e-02  1.81511161e-03
  -4.18561338e-03  2.72806095e-03  1.84774669e-03  2.32571224e-02
  -3.62586866e-03 -2.57413497e-02  4.02977372e-02  5.35851306e-03
  -3.06822235e-02  3.53570097e-02 -1.26496679e-02  5.80675783e-03
   4.60161659e-03 -1.23862180e-02  1.55171471e-02 -2.03109857e-02
   5.36861934e-03  1.44253929e-02 -1.55582018e-02 -3.00789639e-03
   7.09780537e-03  3.78598738e-03 -1.90397802e-03 -4.36166784e-03
  -1.61494752e-02 -9.57092816e-03  1.66469596e-02  2.02867846e-02
   8.74064562e-03 -3.69635089e-02  2.45981688e-02  2.44762415e-05
   2.65297612e-02 -5.65507611e-02  2.11100448e-02 -1.59385542e-02
   2.16523750e-02  8.24320396e-03 -1.19107432e-02 -2.65382701e-03
  -3.12885988e-02 -8.39971804e-04 -5.09638069e-02  3.70207971e-02
  -3.86831154e-03  7.34583067e-03 -7.80391160e-03  2.16666323e-02
   1.44281827e-02  2.34461164e-02  1.89871264e-02  2.49997453e-02
  -4.24011799e-02  3.00994599e-03  1.13842015e-03 -1.36375829e-02
  -2.86653979e-02  2.54946027e-02 -1.41063536e-02 -1.21862282e-03
   4.48972747e-03  6.65803173e-03 -2.69789800e-03 -4.13691953e-03
   2.13004642e-02 -1.60669410e-02 -9.51533592e-03 -5.01331052e-02
   7.61613960e-03 -1.04469781e-02 -7.31934059e-04 -2.81985342e-04
   2.19386377e-03  3.79455409e-03 -1.11748456e-03 -6.00952125e-03
  -9.91363845e-03 -5.09548564e-03 -2.44983072e-02 -1.72975672e-02
   9.50968844e-03 -3.62841711e-02 -4.55958794e-03  1.37125824e-03
   7.18545900e-03 -1.58720685e-03 -1.44475532e-02 -2.23748903e-02
  -4.05693727e-02  3.55632400e-02 -1.30620582e-03  5.38951802e-03
  -1.91137097e-02 -4.27591050e-02 -2.73569197e-02  1.27168276e-03
  -1.85378351e-02 -7.01738243e-04 -2.83759078e-02  8.08597820e-04
  -9.58997930e-03  1.23761410e-03  4.34167906e-03 -6.78296315e-03
   3.57030486e-02 -9.62271107e-03 -9.67085263e-03  0.00000000e+00
   3.24178422e-04  3.26201083e-02 -7.06086075e-02  1.30559437e-02
  -9.62224911e-02  1.89523982e-03  7.78398371e-03  4.54877039e-03
   5.78081958e-03  7.11326511e-03  4.15156887e-04 -1.91974270e-02
  -7.98532087e-03  2.41789074e-02 -4.11431639e-03  1.28425645e-03
  -7.57808090e-03  1.82470266e-02  3.62114623e-02  1.91141494e-02
   2.35279337e-02 -7.22572723e-04 -1.88539069e-03 -4.12193347e-02
   1.60796818e-02  1.12383544e-02  1.21170106e-04  4.12750278e-03
   2.16607546e-03 -1.66681897e-02  2.23429981e-03 -7.14857241e-03
  -4.52925993e-05  6.66764106e-03  6.24784033e-03 -2.78202903e-02
   1.58978262e-02  3.49680312e-03 -6.48520731e-02 -5.65323893e-02
  -1.90193092e-03  3.01207267e-02 -1.31315451e-03  9.92434559e-06
  -6.51746958e-04  1.95772688e-03  1.06466202e-02  8.25969551e-05
  -9.57170886e-03 -2.20494724e-04  9.06388435e-03  2.11596415e-02
   2.60474656e-02  2.26263006e-04  4.14867414e-03  9.75519473e-03
   1.56634480e-03  3.96228420e-03  4.81861835e-03 -1.34948169e-02
  -5.77011371e-03  1.29384089e-03  1.19237926e-02 -1.14615709e-02
  -8.75009192e-03  2.14285082e-02 -4.17197412e-02 -2.44163879e-02
  -3.26237878e-02 -5.81602061e-03 -8.90697174e-03  3.72281689e-03
  -1.25739551e-02 -2.37849114e-02 -1.89637209e-02 -4.30928889e-02
  -2.82602474e-02  1.10434587e-02 -1.18333579e-02  7.62912600e-03
   3.36174950e-03 -9.91674983e-03 -3.36842520e-03  6.13852406e-03
  -1.23463397e-02  2.70532920e-02  1.34658473e-03  2.58183749e-02
  -5.77527036e-03  5.68023344e-02 -9.19399529e-03  1.00286985e-03
   7.84130126e-03  0.00000000e+00 -3.01124948e-03  7.41680648e-03
  -2.78523695e-02 -2.16936666e-03  1.14407793e-03 -4.08355329e-02
   1.12714479e-03  1.00583568e-02 -2.67619252e-02  7.75870591e-03
   2.04462887e-02  1.39865985e-02  1.81644059e-02 -1.93742093e-02
   1.23097657e-02  2.59880695e-02  5.58663393e-04 -3.45414041e-03
   2.05564822e-02 -1.15355666e-02  1.41612339e-03  6.09386922e-03
  -1.14223166e-03 -6.27691093e-03  1.57887123e-03  1.56785695e-02
  -8.30917111e-03  5.98673265e-03 -1.42472019e-02 -4.01584093e-03
   8.80469174e-02  6.22115845e-03  2.91506854e-04 -6.94961622e-04
   3.03143045e-03  2.98517155e-04  3.92965342e-03  7.57468421e-03
   1.36820283e-02 -4.76946092e-03  6.67322670e-03 -2.88096241e-02
  -2.81311951e-02  9.81076193e-03 -2.15307441e-04  3.47471791e-04
   5.49986819e-03  3.01861022e-03 -2.62180172e-03 -4.43657911e-02
  -2.38224758e-02  8.25245727e-03 -1.44065480e-02 -2.50050522e-02
  -7.96490470e-03  1.67455640e-02  1.66711673e-02  7.48092760e-04
   1.81851874e-04  4.37139413e-02 -5.55381346e-04 -2.66707086e-03
  -2.66720264e-02  1.24151910e-02 -5.70328785e-03 -2.67261397e-02
   1.00455909e-03  1.49863493e-02  3.71619172e-03 -7.57665783e-03
  -1.82723099e-03 -8.07269119e-03  2.34465032e-02  1.33758856e-02
   8.01502638e-03 -1.85376857e-03  7.79391675e-03  5.96944686e-03
  -1.88244893e-03 -3.19623791e-03 -2.55225452e-02  1.51594674e-03
   1.68299266e-03  2.77837254e-03  9.18093058e-03 -7.20618269e-03
   2.63524670e-02 -2.19232243e-02  2.98468545e-02  3.63779680e-02
   3.28010985e-02  4.33930745e-03 -1.08360486e-02 -4.29211655e-04
  -1.37964806e-02  0.00000000e+00  2.40683045e-02  2.07455480e-02]]
intercept: [-0.32916385]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_3.pickle')
test_set.head()

# test classifier
predic = best_model.predict(test_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
print("predictions:\n", predic)

predictions:
 [1 1 1 ... 1 1 1]

set(predic)

{0, 1}

# evaluate classifier

print("Report for Support Vector Machine:")
print(_classification_report(test_set['likes'], predic))

print("Accuracy for Support Vector Machine:", _accuracy_score(test_set['likes'], predic)*100)

Report for Support Vector Machine:
              precision    recall  f1-score   support

           0       0.70      0.36      0.48     50930
           1       0.74      0.92      0.82    103063

    accuracy                           0.74    153993
   macro avg       0.72      0.64      0.65    153993
weighted avg       0.73      0.74      0.71    153993

Accuracy for Support Vector Machine: 73.72737721844499

# Confusion matrix for SVC

print("Confusion Matrix for SVC: ")
_confusion_matrix(test_set['likes'], predic)

Confusion Matrix for SVC before balance the data:

array([[18347, 32583],
       [ 7875, 95188]], dtype=int64)

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], predic)

_plt.plot(fpr,tpr)
_plt.xlim([0.0,1.0])
_plt.ylim([0.0,1.0])

_plt.title("SVM ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_jl.dump(best_model, "../models/best_SVM.joblib")

['../models/best_SVM.joblib']

_del_all()

6.2 Random Forest Classifier¶

(see the docs)

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')
train_set.head()

sub_train_set = train_set[:round(train_set.shape[0]/2)]
#sub_train_set = train_set
del train_set
sub_train_set.shape

(279193, 817)

random_forest = _RandomForestClassifier(n_jobs = -1, random_state = 0)
random_forest.get_params()

{'bootstrap': True,
 'class_weight': None,
 'criterion': 'gini',
 'max_depth': None,
 'max_features': 'auto',
 'max_leaf_nodes': None,
 'min_impurity_decrease': 0.0,
 'min_impurity_split': None,
 'min_samples_leaf': 1,
 'min_samples_split': 2,
 'min_weight_fraction_leaf': 0.0,
 'n_estimators': 'warn',
 'n_jobs': -1,
 'oob_score': False,
 'random_state': 0,
 'verbose': 0,
 'warm_start': False}

# fine tune classifier

param_grid = {
    'bootstrap': [True, False],
    'max_depth': [10, 30, 50],
    'min_samples_leaf': [1, 2, 4],
    'min_samples_split': [2, 5, 10],
    'n_estimators': [200, 500, 1000],
    'criterion': ['gini', 'entropy']}
 
grid = _GridSearchCV(estimator=random_forest, param_grid=param_grid, refit=False, verbose=5, cv=3, error_score=_np.nan, n_jobs=-1, pre_dispatch=6)
grid.fit(sub_train_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), sub_train_set['likes'])

Fitting 3 folds for each of 324 candidates, totalling 972 fits

[Parallel(n_jobs=-1)]: Using backend LokyBackend with 12 concurrent workers.
[Parallel(n_jobs=-1)]: Done  12 tasks      | elapsed: 25.0min
[Parallel(n_jobs=-1)]: Done  66 tasks      | elapsed: 134.9min
[Parallel(n_jobs=-1)]: Done 156 tasks      | elapsed: 558.6min
[Parallel(n_jobs=-1)]: Done 282 tasks      | elapsed: 1123.2min
[Parallel(n_jobs=-1)]: Done 444 tasks      | elapsed: 2490.3min
[Parallel(n_jobs=-1)]: Done 642 tasks      | elapsed: 3490.7min
[Parallel(n_jobs=-1)]: Done 876 tasks      | elapsed: 4913.4min
[Parallel(n_jobs=-1)]: Done 972 out of 972 | elapsed: 5701.6min finished

GridSearchCV(cv=3, error_score=nan,
             estimator=RandomForestClassifier(bootstrap=True, class_weight=None,
                                              criterion='gini', max_depth=None,
                                              max_features='auto',
                                              max_leaf_nodes=None,
                                              min_impurity_decrease=0.0,
                                              min_impurity_split=None,
                                              min_samples_leaf=1,
                                              min_samples_split=2,
                                              min_weight_fraction_leaf=0.0,
                                              n_estimators='warn', n_jobs=-1,
                                              oob_score=False, random_state=0,
                                              verbose=0, warm_start=False),
             iid='warn', n_jobs=-1,
             param_grid={'bootstrap': [True, False],
                         'criterion': ['gini', 'entropy'],
                         'max_depth': [10, 30, 50],
                         'min_samples_leaf': [1, 2, 4],
                         'min_samples_split': [2, 5, 10],
                         'n_estimators': [200, 500, 1000]},
             pre_dispatch=6, refit=False, return_train_score=False,
             scoring=None, verbose=5)

print("best params:\n", grid.best_params_)
print("best score:", grid.best_score_)

best params:
 {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 1000}
best score: 0.7453481999906875

with open("../models/random_forest_params.txt", 'w') as f:
    f.write(str(grid.best_params_))

print("results:", grid.cv_results_)

results: {'mean_fit_time': array([ 252.87800407,  600.7321078 , 1147.74082383,  311.25228389,
        597.79628388, 1124.06511203,  307.2705942 ,  662.01461442,
       1102.80993398,  303.65359688,  676.63054156, 1121.13627799,
        290.6686426 ,  672.74625977, 1103.33752354,  299.24372005,
        666.57276177, 1089.51547178,  310.32110922,  677.31205535,
       1087.89945881,  302.18119884,  681.54606994, 1077.58835777,
        343.88937259,  647.70520091, 1116.92719674,  748.18958219,
       1670.8926479 , 3277.3076601 ,  726.21665557, 1612.23545631,
       3128.41236226,  706.48307268, 1692.42808255, 3044.78944604,
        710.94016171, 1687.92898154, 3029.70489526,  716.6063451 ,
       1661.93725832, 2975.88908982,  699.15533026, 1650.08494091,
       2936.43123945,  663.98700976, 1589.69072064, 2841.49037019,
        677.26351619, 1600.60654235, 2818.42968504,  689.23385302,
       1575.95244972, 2845.27126567,  854.03630026, 1924.91824579,
       3589.90236084,  806.29059982, 1901.73787904, 3542.07222215,
        755.210814  , 1842.24359202, 3356.92748769,  772.36362791,
       1820.98641507, 3347.10009464,  775.59133355, 1795.96330682,
       3266.85892995,  748.71085684, 1781.311143  , 3273.03043238,
        728.07022198, 1694.15878701, 3058.85896428,  699.67593479,
       1671.264992  , 2974.15439741,  733.93203028, 1685.07307522,
       2878.58820073,  316.39686712,  704.58548148, 1315.41525666,
        266.52052689,  617.80446243, 1116.14761202,  323.40379898,
        596.31457845, 1192.99234486,  306.07180023,  631.73588936,
       1107.42891963,  297.97278309,  667.9530731 , 1098.42698288,
        303.02261813,  669.46902156, 1087.44833231,  303.09010363,
        673.5451235 , 1114.85340317,  306.89094249,  687.86916415,
       1097.86382119,  297.0615526 ,  674.32869554, 1090.86054126,
        759.74303031, 1654.02773229, 3203.89591153,  739.06663569,
       1612.05327415, 3126.06397398,  717.29151392, 1664.62706677,
       2990.3717099 ,  713.38628578, 1674.57015268, 3052.98233859,
        706.96910882, 1662.88572113, 2994.57327501,  696.03201071,
       1641.71012592, 2944.10638984,  676.90547228, 1583.02720237,
       2877.37689845,  686.42369564, 1584.088492  , 2868.15209929,
        685.93400542, 1559.20022694, 2867.99252415,  829.33399868,
       1916.37242405, 2354.95233067,  793.30663912, 1802.85570685,
       2807.89743249,  762.03286791, 1283.57182018, 3463.58312575,
        731.26084558, 1854.44696649, 3406.09970792,  775.27916543,
       1776.74966931, 3378.01250974,  751.70151536, 1781.36533038,
       3312.62044843,  715.7409931 , 1679.43034275, 3086.08285077,
        707.93253167, 1689.3303593 , 3104.21390899,  733.89479637,
       1658.40256365, 3009.73295474,  387.77073312,  891.25081325,
       1669.05788708,  303.89761273,  792.31527678, 1507.48082161,
        375.99388091,  780.54043563, 1485.7558972 ,  384.29701686,
        844.19593382, 1433.76853601,  375.6255339 ,  863.93716296,
       1459.30460747,  373.0131824 ,  874.46136022, 1458.84884469,
        366.60730712,  847.71785371, 1464.13846231,  386.84738477,
        851.90233604, 1431.67779454,  376.84327952,  868.41485691,
       1490.45787962, 1037.9270467 , 2405.62106681, 4825.63027302,
       1033.44868628, 2339.84804352, 4577.08206526, 1021.59072081,
       2459.0623556 , 4526.76158404, 1010.90661224, 2410.19199936,
       4414.15627249,  985.28743219, 2401.53081822, 4428.08470591,
        977.47797553, 2381.53228196, 4365.35772228,  952.73910848,
       2311.57463264, 4176.69107254,  959.27862748, 2286.79719615,
       4208.51034379,  948.67896342, 2275.09348027, 4208.80089919,
       1235.81870461, 2858.33534185, 5524.50904346, 1157.73344286,
       2814.34227665, 5262.66301656, 1120.64758428, 2752.72798594,
       5050.97557322, 1114.87002667, 2681.47512833, 5052.0563302 ,
       1080.7848107 , 2693.51625339, 4898.46493554, 1077.82605847,
       2566.85985637, 4847.68583147, 1029.18784451, 2430.99676498,
       4506.637381  , 1028.56274716, 2459.17040197, 4529.50425267,
       1022.89356637, 2430.04791927, 4297.7135512 ,  366.22898634,
        874.93709048, 1706.81329211,  423.63784893,  806.94950231,
       1431.94374752,  387.1473972 ,  824.15550756, 1517.69511525,
        388.56926513,  841.67227936, 1471.32414134,  372.09363929,
        845.24180651, 1432.31375853,  381.16140246,  855.78428976,
       1459.6618127 ,  386.32925193,  868.99546496, 1440.26550515,
        388.21920252,  858.20249327, 1467.01532388,  370.12357601,
        863.297539  , 1474.60670225, 1016.42951632, 2406.92572657,
       4735.78016353, 1007.15231657, 2369.1623466 , 4589.57932417,
        996.3950723 , 2456.83763719, 4445.20094522, 1017.31780783,
       2411.83959389, 4458.39833736,  999.47517347, 2423.5928425 ,
       4455.27136358,  983.52447764, 2379.59146659, 4360.11706217,
        971.04184556, 2321.9312737 , 4224.82240629,  949.03002222,
       2311.84523551, 4232.07036241,  954.30591869, 2300.70933517,
       4256.19653241, 1227.57042034, 2862.16809591, 5481.17687869,
       1156.52833215, 2833.5944798 , 5331.72757514, 1125.75044847,
       2745.76227482, 5013.83818905, 1138.77612058, 2735.31786243,
       5126.30403113, 1114.51132131, 2684.64896623, 4955.564562  ,
       1108.3780485 , 2676.21811597, 4920.55986961, 1029.19738499,
       2504.03147769, 4686.9357365 , 1041.30135949, 2464.48220364,
       4656.1280938 , 1026.20438417, 2470.45756213, 2600.50956917]), 'std_fit_time': array([ 30.30829678,   7.3920137 ,   5.36162458,   6.16287781,
         4.71603199,  13.29157039,  30.77081708,  24.55500663,
         2.9462911 ,   8.21775426,  22.41828531,  17.96219934,
        14.42042926,   3.55636399,   3.40924981,  28.19064984,
        29.65923753,   9.68237659,  40.21162156,  26.81535584,
         2.73972701,  40.72857279,   5.32890289,   9.04349199,
        72.35007697,  33.76382866,  12.70825532,   1.1973388 ,
        26.28799272,  27.57452577,  19.2399905 ,  33.57233439,
         7.03525377,   6.96787114,  26.17840237,  21.69617215,
         5.72652778,  16.09784563,   5.60459622,  38.07846644,
         5.13522422,   6.10484057,  28.8840013 ,   9.15673828,
         3.77868333,  14.54373134,  22.62081667,  42.12137488,
        46.36612411,  48.95809511,   3.71756459,  10.25182197,
        38.77469533,   9.91279303,  15.02808539,   3.50673732,
        21.2591932 ,  33.73133942,  22.16604837,  27.53982584,
         6.55108126,  15.15851592,  29.02630199,  13.98807251,
        15.6170111 ,   5.27449468,   4.71669523,   4.94979628,
         8.21982903,  28.0828732 ,   6.64810862,  10.47285877,
        20.08594434,  21.47197092,  42.8724304 ,  13.53014429,
        26.3550897 ,  16.34642299,  23.8984927 ,  35.45706558,
         9.59644955,   5.7351674 ,  13.51277839,  18.27234975,
        31.81341328,   5.88774603,   4.85636529,  42.75257169,
        18.926562  ,  51.05243492,  14.43683902,  20.65098564,
        11.84923354,   5.44938135,  21.60465023,  11.25836454,
        24.45109409,  25.05591148,   8.4204074 ,  46.71867588,
        59.42065962,  16.06614626,   7.70801151,  17.29630675,
         4.24164931,  29.93605116,   9.75911225,  19.15552767,
        59.62827848,  13.51343739,  27.27137587,   6.4782706 ,
         3.12380134,  20.32793836,   6.62062667,  24.18534535,
        40.09660025,  49.38004415,  15.97957569,  26.38353192,
        14.78771054,  22.02305526,  21.94410114,  29.71031024,
        11.65456113,  14.25976303,  39.85181605,  10.32180083,
        37.968235  ,  33.46566647,  17.14615515,  33.27546592,
        13.19411005,  17.38057552,   6.11399812,  21.17612706,
        28.40688046,  51.34668928,  12.63107768,  62.71129915,
         0.36834374,   4.49424576,   4.33024111,   2.13752287,
         3.14354585,   2.83718275,   5.25167961,   1.73576136,
         1.3772604 ,   6.43414514,   5.18661129,   4.81210139,
         4.70039194,   3.52572648,   9.0589487 ,   3.79769031,
        17.04920476,  10.70438956,  15.14760422,  21.8990582 ,
         7.90116057,   7.4321548 ,  23.49232366,   8.12951472,
        11.25133362,   8.04122486,  22.22850683,   4.90010178,
        25.5742791 ,  12.17891482,   4.51435936,   8.93586555,
        15.36015864,   1.89146275,  18.4743522 ,  25.47284129,
        21.10220944,  26.36448207,  19.62003341,   0.5027232 ,
        20.74194503,   6.88672995,   5.34562602,  17.03257041,
        33.73822554,   6.77362179,   3.55545009,  27.0746333 ,
        18.11377592,  22.28935143,   8.96315801,  12.64294647,
         8.56096107,  12.50241136,  11.83272088,   7.44797562,
        56.87594953,  29.40598311,  10.79544944,  15.02910549,
        10.99092638,  20.06998157,  21.94652134,   3.90372069,
        16.31665264,  28.43944401,   2.07436933,  20.84868541,
         7.52533085,  20.92857139,   6.96924701,  12.79672038,
        17.67013584,  16.1245658 ,  15.55423661,  21.97211009,
        35.49759124,  30.27700544,   8.69162715,  12.27562495,
        22.06539704,  60.08812146,  14.55264842,  40.72960289,
        91.0545018 ,  11.58906563,   9.32032197,  30.75423376,
        10.83326925,  34.30339473, 112.39188405,  30.24681146,
         3.66167295,  31.26381741,  20.76093123,  12.31057523,
        39.19107482,  32.69729193,  14.48249089,  17.78150583,
        26.80245948,  13.89638986,  31.43113123,   6.77775749,
        14.62908714,  29.0209967 ,  26.23457069,  28.85436514,
        23.38462071,  20.3009723 ,   9.2764942 ,  39.71354787,
        18.69865738,  42.06633349,  19.12809447,  18.12269559,
        17.00438512,   2.51240561,  37.01248076,  17.14268623,
        20.883293  ,   1.64621857,   2.43930551,   1.74417196,
        51.9362573 ,  17.33219577,  38.648948  ,   0.62063819,
        14.20026948,   3.15050465,  29.67027009,  22.6174495 ,
        46.84971878,   4.36123567,  39.84663268,  17.87730277,
        12.26934706,  12.82897682,  29.01803997,  22.89018772,
        20.59199815,   1.00230024,   4.11090258,   7.74640375,
        28.20557553,  14.72147182,  25.71679854,  17.53483287,
        24.18787215,  18.48195342,   9.32786236,   3.19651885,
        20.79966976,  19.34164785,   5.23871617,  35.98229101,
         9.04758183,  17.21375522,   5.52598687,  12.11320323,
         3.80078647,   7.92999699,  12.25112275,   8.95235647,
         7.91797862, 233.30396124,  10.83876671,   5.30819944,
        16.55791144,  58.16967771,  15.74847618,  60.70317638,
         9.28805116,   3.96437891,   9.86981749,  26.12548248,
         7.69346146,  41.7562853 ,  15.55690244,  12.77329849,
         7.44098945,  35.39946121,  13.85279007, 266.83686164]), 'mean_score_time': array([ 22.74486287,  27.54935249,  40.42393446,  20.96096595,
        24.16673144,  33.61048714,  24.09159581,  21.4885548 ,
        27.27940639,  21.48090823,  23.90609399,  40.81389364,
        22.29373638,  25.46559215,  30.36516023,  21.57100264,
        21.2904187 ,  28.73950942,  20.99520882,  26.80766948,
        31.47818518,  22.31800421,  25.20029982,  34.49511989,
        20.40345565,  27.27109845,  47.96943299,  38.38139757,
        75.43567824, 142.1543235 ,  36.46684798,  84.20955404,
       117.19570931,  39.00838955,  66.25654753, 124.20829352,
        33.95057321,  71.15013234, 113.59234134,  38.96616999,
        67.43705702, 128.9812034 ,  32.80895972,  62.08403794,
       104.8068277 ,  33.91367467,  55.18680604, 103.22472239,
        33.0865519 ,  53.52890388,  91.599799  ,  34.16134564,
        52.56714249, 111.06709154,  43.28462021,  92.08084838,
       204.00564702,  43.39299862,  89.72115509, 156.64757824,
        38.93026106,  79.38644799, 154.51361744,  44.90229925,
        77.89443612, 134.62910597,  44.05755647,  76.60754387,
       148.97974277,  39.98810156,  76.71492179, 124.91507316,
        36.80760376,  66.35561681, 106.63028669,  39.01138043,
        64.5846839 , 104.55516775,  34.45987995,  64.06706834,
        98.58512616,  29.66336902,  32.21454899,  48.465108  ,
        22.88017066,  28.25613244,  41.74274349,  22.84559528,
        30.67034086,  43.45051614,  22.64645966,  24.40409335,
        35.75275524,  20.08098539,  22.798721  ,  31.6739943 ,
        22.74087501,  29.15207044,  38.58485301,  32.19127655,
        21.82200146,  29.69894179,  19.80871503,  21.07599092,
        30.05299664,  23.85722311,  27.23353243,  47.80221279,
        37.50374508,  83.11082554, 131.89508184,  37.60214726,
        80.8807865 ,  97.29390987,  37.50208131,  64.13488857,
       125.63415074,  36.20155899,  64.83169134, 103.95942688,
        37.98412506,  66.09464622, 121.3236746 ,  34.38740889,
        60.95438727, 103.46906996,  36.68692787,  53.5830938 ,
        98.62202724,  35.92562858,  56.79749886,  92.94353922,
        33.38409058,  54.9029003 , 105.8390673 ,  43.9564929 ,
        90.96881978,  74.26267052,  35.22018433,  81.99613849,
        69.53909079,  30.86548845,  19.91211637, 165.69738777,
        34.17664115,  79.06464195, 131.44495098,  34.8707815 ,
        48.8439579 , 159.30879815,  31.32160385,  59.16483959,
        77.00315078,  28.22920283,  38.79462775, 117.50687774,
        30.0519975 ,  51.31648493,  71.80637836,  32.39673042,
        36.52602466,  71.91708016,  26.34690038,  28.0852557 ,
        27.46225333,  20.06535864,  28.58791248,  39.13438503,
        25.13347801,  26.39776413,  33.74811896,  22.44765735,
        20.03377787,  36.5669144 ,  21.17771935,  28.27840439,
        32.7019105 ,  21.15378459,  23.29040798,  40.95878116,
        21.03643068,  26.94464   ,  31.16934427,  22.67404548,
        24.22856458,  24.4719158 ,  19.63318054,  24.31400243,
        55.21044834,  36.6776185 ,  81.42533151, 173.99586908,
        40.21050549,  90.37806646, 150.76397332,  39.94188897,
        73.09294057, 138.47282974,  40.28896332,  73.73821322,
       131.70758104,  38.85878762,  80.82327445, 140.61676288,
        39.93557556,  67.3748912 , 116.77915653,  37.2760191 ,
        69.67506965, 117.97761822,  37.90533606,  68.95566368,
       120.58597914,  36.66299057,  68.35793169, 111.96735199,
        52.30983059, 115.45236993, 249.59377821,  52.13097493,
       109.56909823, 223.20033431,  49.86469928, 114.73362462,
       175.27345332,  40.55359038,  94.66859619, 207.41952109,
        45.88666828, 105.61533038, 179.65640362,  40.31389523,
        87.28633006, 147.3414553 ,  38.84282859,  76.10887901,
       143.36608315,  40.13105273,  77.18898861, 119.65613238,
        41.25371949,  75.5912625 , 127.92535814,  23.59692009,
        31.74813008,  48.87002476,  26.67868336,  36.05129306,
        45.2753044 ,  25.54803848,  30.76608968,  38.37342095,
        21.76381874,  27.10653933,  34.0479823 ,  22.49087803,
        26.45361725,  35.12310855,  22.54938698,  27.99050943,
        35.33287899,  17.09662922,  19.62021629,  25.16173673,
        21.79473607,  26.04038612,  35.8611358 ,  18.33332284,
        24.99484897,  40.30558395,  39.98743526,  81.91469073,
       161.06310924,  40.43723281,  81.71322838, 137.18660172,
        39.01636712,  74.91539947, 129.77674564,  34.74910665,
        70.54807472, 141.5206809 ,  39.8634367 ,  73.80270783,
       130.4791983 ,  36.25807516,  65.96432996, 120.82633654,
        36.08719714,  62.1285855 , 123.06800938,  33.53169688,
        58.39689056, 111.50492374,  33.91101575,  61.79248222,
       122.0011967 ,  45.3614041 , 116.23760295, 242.4129866 ,
        48.1316669 , 115.09100095, 232.28936831,  46.3434461 ,
       129.92767119, 185.7609218 ,  42.03429993, 101.12400738,
       192.25539096,  46.57715472, 102.40990162, 170.98325523,
        36.91997067,  87.0579412 , 145.6981825 ,  38.12441715,
        75.73321358, 130.70625877,  38.28798119,  66.29643885,
       118.93605439,  43.10410603,  74.20729534,  34.68228738]), 'std_score_time': array([ 3.73790697,  0.30415656,  6.92587549,  2.80850105,  1.15761239,
       11.03193416,  2.18269738,  5.7843071 ,  6.68902138,  4.04390138,
        6.90222302,  7.87328813,  4.49783489,  2.72922677,  7.98650193,
        1.52263762,  5.5064888 ,  3.48344337,  0.47871017,  3.73582713,
        4.34613959,  2.36834954,  1.46729828,  5.89990151,  1.61809314,
        3.84808417,  5.0427848 ,  2.30684875,  4.81018346, 10.27616733,
        2.74101451,  3.79907696,  3.98529257,  4.40175377,  2.61252675,
        5.53625533,  3.37931338,  1.65719739,  1.86480187,  0.65404252,
        5.25289516,  4.36917093,  3.74452769,  4.16126799,  5.23263841,
        3.13796786,  7.31170198,  7.82135388,  1.95445351,  4.38168262,
       10.20791492,  3.05717731,  3.1417052 ,  6.26757503,  4.58113999,
       11.0739803 ,  6.86371335,  4.36039996,  2.19959567,  1.47891058,
        2.45664485,  3.11531965,  1.32099943,  4.53794374,  8.37548576,
        9.20223396,  4.9506376 ,  1.53459068,  0.28753037,  2.98474871,
        3.14313634,  1.19331635,  2.89314709,  0.77832221,  4.2531609 ,
        2.10041339,  4.41834301,  5.39228948,  3.26462031,  2.35477828,
        3.51190062,  9.50254451,  0.72167435,  1.73167494,  2.07519971,
        4.71073193,  2.34675829,  2.15203514,  3.67366729,  6.62659062,
        2.37555739,  2.51017608,  7.64574316,  2.75829883,  0.63760264,
        4.81559447,  1.73884145,  4.53266327,  5.38363662, 10.35304377,
        4.83179403,  7.91551107,  1.98318509,  5.88545707,  2.44557327,
        1.27360577,  3.45900966,  5.74310212,  1.51303855,  3.9097375 ,
       10.87698257,  4.39227344,  8.45290361,  9.14867945,  5.90746222,
        5.87062798,  5.80196265,  2.26114274,  0.93526402,  6.58166393,
        0.46877206,  6.95247138, 12.93359514,  1.36948163,  4.16947603,
        5.69149097,  1.43797017,  9.73990258,  2.87744378,  3.04764653,
        5.66657796,  6.6383046 ,  1.88310248,  4.53918743, 14.56719218,
        4.5464202 ,  3.90785792, 16.59522595,  5.77814509, 13.45240866,
        6.98776699,  3.54062851,  0.12516608,  0.24607868,  0.66054981,
        5.3054573 ,  2.77873505,  1.25965318,  4.11532532,  2.06323342,
        1.99657908,  1.66333947,  4.66710504,  0.55396838,  6.17189624,
        3.75059384,  3.96108726,  4.51974468, 19.41740517,  3.72227094,
       15.4103076 , 12.85118308,  2.91415565,  2.36420569,  7.39524787,
        1.42865041,  7.54756348,  5.08740656,  4.49961882,  2.20533559,
       12.30644421,  2.95896234,  8.56015294,  9.45873904,  0.76678208,
        0.8716595 ,  6.08612062,  2.98883532,  6.7323311 ,  3.27947057,
        0.14285844,  4.45707189,  6.15440368,  1.70068028,  4.83515932,
        5.03018124,  2.97086614,  5.4358429 ,  2.0888408 ,  3.05860615,
        7.75705428,  4.24975645,  4.34786636,  4.31404968,  4.11678185,
        1.75035116,  3.61430526,  0.45768793,  7.25933701,  2.53974597,
        9.3456339 ,  2.07941927,  3.03685232,  7.37798915,  4.90112696,
        6.08297988,  5.68544687,  3.41365848,  5.19504503,  6.27730193,
        4.02094883,  2.10016744, 10.31717127,  2.40784982,  3.76999755,
       10.16835344,  3.02120987,  1.6133806 , 17.22251422,  5.18892895,
        4.67723033, 13.7662241 ,  7.40142595, 13.31781516,  9.85166008,
        5.56724033,  7.27027481, 21.16881461,  4.52884724, 20.45550601,
        4.03839222,  3.6416739 ,  3.15661309,  7.46747975,  3.01372365,
        5.64001772,  3.40005478,  3.07717959,  3.0524081 ,  5.48947086,
        2.04908638,  3.11329794,  3.7344732 ,  0.97981083,  2.09441675,
        4.67694722,  1.61361817,  5.13391137,  2.10023593,  0.77977137,
        1.35841737,  6.64283177,  1.23429067,  2.70003199,  6.35901372,
        2.15777896,  6.75359925,  6.36054868,  1.70093369,  3.16029066,
        5.18878811,  0.54288971,  0.31328513,  2.14584227,  2.06049559,
        6.19079603,  8.08457458,  1.53124759,  3.72749796,  1.61386877,
        4.85124001,  2.60318048, 21.41306405,  0.44808795,  1.72836091,
        1.57733826,  1.68303042,  7.2501898 ,  7.81201401,  3.84858105,
        7.92796487,  4.47742288,  2.39859737,  2.52222502,  6.60925813,
        3.56953739,  5.61733233,  1.80074324,  2.59539304,  6.54945999,
        9.54982783,  5.54282533,  8.48413645,  6.77137514,  2.65786256,
        6.28151036,  6.16077712,  3.38522252,  7.16557433, 15.11352743,
        0.61040563, 16.13892246,  9.48657747,  9.43038419, 25.17788277,
        0.63656661,  2.00818815, 14.88494676, 13.4548406 ,  0.60568569,
       17.64370399,  5.56989484,  5.03281369,  2.0426668 ,  5.30472464,
        0.98666712,  5.50817538, 17.00630192,  1.30159059, 11.35127102,
        5.20120321,  2.80094782,  5.02332418, 16.61052917]), 'param_bootstrap': masked_array(data=[True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   True, True, True, True, True, True, True, True, True,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False],
             mask=[False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False],
       fill_value='?',
            dtype=object), 'param_criterion': masked_array(data=['gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'gini', 'gini', 'gini', 'gini',
                   'gini', 'gini', 'gini', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy', 'entropy',
                   'entropy', 'entropy', 'entropy', 'entropy'],
             mask=[False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False],
       fill_value='?',
            dtype=object), 'param_max_depth': masked_array(data=[10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
                   10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 30,
                   30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30,
                   30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 50, 50,
                   50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50,
                   50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 10, 10, 10,
                   10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
                   10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 30, 30, 30, 30,
                   30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30,
                   30, 30, 30, 30, 30, 30, 30, 30, 30, 50, 50, 50, 50, 50,
                   50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50,
                   50, 50, 50, 50, 50, 50, 50, 50, 10, 10, 10, 10, 10, 10,
                   10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
                   10, 10, 10, 10, 10, 10, 10, 30, 30, 30, 30, 30, 30, 30,
                   30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30,
                   30, 30, 30, 30, 30, 30, 50, 50, 50, 50, 50, 50, 50, 50,
                   50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50,
                   50, 50, 50, 50, 50, 10, 10, 10, 10, 10, 10, 10, 10, 10,
                   10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
                   10, 10, 10, 10, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30,
                   30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30,
                   30, 30, 30, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50,
                   50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50,
                   50, 50],
             mask=[False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False],
       fill_value='?',
            dtype=object), 'param_min_samples_leaf': masked_array(data=[1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2,
                   4, 4, 4, 4, 4, 4, 4, 4, 4, 1, 1, 1, 1, 1, 1, 1, 1, 1,
                   2, 2, 2, 2, 2, 2, 2, 2, 2, 4, 4, 4, 4, 4, 4, 4, 4, 4,
                   1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2,
                   4, 4, 4, 4, 4, 4, 4, 4, 4, 1, 1, 1, 1, 1, 1, 1, 1, 1,
                   2, 2, 2, 2, 2, 2, 2, 2, 2, 4, 4, 4, 4, 4, 4, 4, 4, 4,
                   1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2,
                   4, 4, 4, 4, 4, 4, 4, 4, 4, 1, 1, 1, 1, 1, 1, 1, 1, 1,
                   2, 2, 2, 2, 2, 2, 2, 2, 2, 4, 4, 4, 4, 4, 4, 4, 4, 4,
                   1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2,
                   4, 4, 4, 4, 4, 4, 4, 4, 4, 1, 1, 1, 1, 1, 1, 1, 1, 1,
                   2, 2, 2, 2, 2, 2, 2, 2, 2, 4, 4, 4, 4, 4, 4, 4, 4, 4,
                   1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2,
                   4, 4, 4, 4, 4, 4, 4, 4, 4, 1, 1, 1, 1, 1, 1, 1, 1, 1,
                   2, 2, 2, 2, 2, 2, 2, 2, 2, 4, 4, 4, 4, 4, 4, 4, 4, 4,
                   1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2,
                   4, 4, 4, 4, 4, 4, 4, 4, 4, 1, 1, 1, 1, 1, 1, 1, 1, 1,
                   2, 2, 2, 2, 2, 2, 2, 2, 2, 4, 4, 4, 4, 4, 4, 4, 4, 4],
             mask=[False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False],
       fill_value='?',
            dtype=object), 'param_min_samples_split': masked_array(data=[2, 2, 2, 5, 5, 5, 10, 10, 10, 2, 2, 2, 5, 5, 5, 10, 10,
                   10, 2, 2, 2, 5, 5, 5, 10, 10, 10, 2, 2, 2, 5, 5, 5, 10,
                   10, 10, 2, 2, 2, 5, 5, 5, 10, 10, 10, 2, 2, 2, 5, 5, 5,
                   10, 10, 10, 2, 2, 2, 5, 5, 5, 10, 10, 10, 2, 2, 2, 5,
                   5, 5, 10, 10, 10, 2, 2, 2, 5, 5, 5, 10, 10, 10, 2, 2,
                   2, 5, 5, 5, 10, 10, 10, 2, 2, 2, 5, 5, 5, 10, 10, 10,
                   2, 2, 2, 5, 5, 5, 10, 10, 10, 2, 2, 2, 5, 5, 5, 10, 10,
                   10, 2, 2, 2, 5, 5, 5, 10, 10, 10, 2, 2, 2, 5, 5, 5, 10,
                   10, 10, 2, 2, 2, 5, 5, 5, 10, 10, 10, 2, 2, 2, 5, 5, 5,
                   10, 10, 10, 2, 2, 2, 5, 5, 5, 10, 10, 10, 2, 2, 2, 5,
                   5, 5, 10, 10, 10, 2, 2, 2, 5, 5, 5, 10, 10, 10, 2, 2,
                   2, 5, 5, 5, 10, 10, 10, 2, 2, 2, 5, 5, 5, 10, 10, 10,
                   2, 2, 2, 5, 5, 5, 10, 10, 10, 2, 2, 2, 5, 5, 5, 10, 10,
                   10, 2, 2, 2, 5, 5, 5, 10, 10, 10, 2, 2, 2, 5, 5, 5, 10,
                   10, 10, 2, 2, 2, 5, 5, 5, 10, 10, 10, 2, 2, 2, 5, 5, 5,
                   10, 10, 10, 2, 2, 2, 5, 5, 5, 10, 10, 10, 2, 2, 2, 5,
                   5, 5, 10, 10, 10, 2, 2, 2, 5, 5, 5, 10, 10, 10, 2, 2,
                   2, 5, 5, 5, 10, 10, 10, 2, 2, 2, 5, 5, 5, 10, 10, 10,
                   2, 2, 2, 5, 5, 5, 10, 10, 10, 2, 2, 2, 5, 5, 5, 10, 10,
                   10, 2, 2, 2, 5, 5, 5, 10, 10, 10],
             mask=[False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False],
       fill_value='?',
            dtype=object), 'param_n_estimators': masked_array(data=[200, 500, 1000, 200, 500, 1000, 200, 500, 1000, 200,
                   500, 1000, 200, 500, 1000, 200, 500, 1000, 200, 500,
                   1000, 200, 500, 1000, 200, 500, 1000, 200, 500, 1000,
                   200, 500, 1000, 200, 500, 1000, 200, 500, 1000, 200,
                   500, 1000, 200, 500, 1000, 200, 500, 1000, 200, 500,
                   1000, 200, 500, 1000, 200, 500, 1000, 200, 500, 1000,
                   200, 500, 1000, 200, 500, 1000, 200, 500, 1000, 200,
                   500, 1000, 200, 500, 1000, 200, 500, 1000, 200, 500,
                   1000, 200, 500, 1000, 200, 500, 1000, 200, 500, 1000,
                   200, 500, 1000, 200, 500, 1000, 200, 500, 1000, 200,
                   500, 1000, 200, 500, 1000, 200, 500, 1000, 200, 500,
                   1000, 200, 500, 1000, 200, 500, 1000, 200, 500, 1000,
                   200, 500, 1000, 200, 500, 1000, 200, 500, 1000, 200,
                   500, 1000, 200, 500, 1000, 200, 500, 1000, 200, 500,
                   1000, 200, 500, 1000, 200, 500, 1000, 200, 500, 1000,
                   200, 500, 1000, 200, 500, 1000, 200, 500, 1000, 200,
                   500, 1000, 200, 500, 1000, 200, 500, 1000, 200, 500,
                   1000, 200, 500, 1000, 200, 500, 1000, 200, 500, 1000,
                   200, 500, 1000, 200, 500, 1000, 200, 500, 1000, 200,
                   500, 1000, 200, 500, 1000, 200, 500, 1000, 200, 500,
                   1000, 200, 500, 1000, 200, 500, 1000, 200, 500, 1000,
                   200, 500, 1000, 200, 500, 1000, 200, 500, 1000, 200,
                   500, 1000, 200, 500, 1000, 200, 500, 1000, 200, 500,
                   1000, 200, 500, 1000, 200, 500, 1000, 200, 500, 1000,
                   200, 500, 1000, 200, 500, 1000, 200, 500, 1000, 200,
                   500, 1000, 200, 500, 1000, 200, 500, 1000, 200, 500,
                   1000, 200, 500, 1000, 200, 500, 1000, 200, 500, 1000,
                   200, 500, 1000, 200, 500, 1000, 200, 500, 1000, 200,
                   500, 1000, 200, 500, 1000, 200, 500, 1000, 200, 500,
                   1000, 200, 500, 1000, 200, 500, 1000, 200, 500, 1000,
                   200, 500, 1000, 200, 500, 1000, 200, 500, 1000, 200,
                   500, 1000, 200, 500, 1000, 200, 500, 1000, 200, 500,
                   1000, 200, 500, 1000],
             mask=[False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False, False, False, False, False,
                   False, False, False, False],
       fill_value='?',
            dtype=object), 'params': [{'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': True, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'gini', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 30, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 1, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 2, 'min_samples_split': 10, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 2, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 5, 'n_estimators': 1000}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 200}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 500}, {'bootstrap': False, 'criterion': 'entropy', 'max_depth': 50, 'min_samples_leaf': 4, 'min_samples_split': 10, 'n_estimators': 1000}], 'split0_test_score': array([0.7255789 , 0.72637404, 0.72668565, 0.72754526, 0.72776017,
       0.72755601, 0.72722291, 0.7275775 , 0.72708322, 0.72700801,
       0.72706173, 0.72741632, 0.72811476, 0.72773868, 0.72743781,
       0.72637404, 0.72760974, 0.72677161, 0.72677161, 0.72709397,
       0.72734111, 0.72677161, 0.72709397, 0.72734111, 0.72747005,
       0.72717993, 0.72705099, 0.74254553, 0.74325471, 0.74348036,
       0.74353409, 0.7445119 , 0.74443668, 0.74258851, 0.74338366,
       0.74428625, 0.7436308 , 0.74355558, 0.74370601, 0.74313652,
       0.7435126 , 0.74324397, 0.74266373, 0.74343738, 0.74345887,
       0.74189008, 0.74271746, 0.74257777, 0.74189008, 0.74271746,
       0.74257777, 0.74314726, 0.74281416, 0.7425133 , 0.73950465,
       0.74107344, 0.74104121, 0.74297534, 0.74328695, 0.74345887,
       0.74382421, 0.74403911, 0.74371676, 0.74382421, 0.74426476,
       0.7440606 , 0.74291087, 0.74334068, 0.7441788 , 0.74377048,
       0.74422178, 0.74432923, 0.74229839, 0.74301832, 0.74309354,
       0.74229839, 0.74301832, 0.74309354, 0.74195455, 0.74286789,
       0.74316875, 0.72595498, 0.72585827, 0.72609467, 0.72626659,
       0.72676087, 0.72707248, 0.72693279, 0.72685757, 0.72738409,
       0.72606243, 0.72683608, 0.72708322, 0.72723365, 0.72669639,
       0.72649224, 0.7271262 , 0.72686832, 0.72632031, 0.72625584,
       0.72662118, 0.72655671, 0.72625584, 0.72662118, 0.72655671,
       0.72729812, 0.72722291, 0.72717993, 0.74312577, 0.74346962,
       0.74421104, 0.74350185, 0.74446892, 0.74389943, 0.74360931,
       0.74402837, 0.74436147, 0.74335142, 0.74423253, 0.74402837,
       0.74321173, 0.74377048, 0.7439639 , 0.74278193, 0.74352334,
       0.74373825, 0.74258851, 0.7432977 , 0.74295385, 0.74258851,
       0.7432977 , 0.74295385, 0.74233063, 0.7427282 , 0.74221243,
       0.74092301, 0.74183635, 0.74147101, 0.74344813, 0.74379197,
       0.74409284, 0.74319024, 0.74450116, 0.74453339, 0.74398539,
       0.7446301 , 0.74455488, 0.74360931, 0.74421104, 0.74453339,
       0.74437221, 0.74428625, 0.74466233, 0.74369527, 0.74356632,
       0.74364154, 0.74369527, 0.74356632, 0.74364154, 0.74237361,
       0.74278193, 0.7432977 , 0.72561113, 0.72627733, 0.72640628,
       0.72668565, 0.72755601, 0.72781389, 0.72641702, 0.72722291,
       0.72786762, 0.72780315, 0.7279106 , 0.72720142, 0.72744856,
       0.72786762, 0.72772793, 0.72765272, 0.72738409, 0.72723365,
       0.72539623, 0.72647075, 0.72681459, 0.72539623, 0.72647075,
       0.72681459, 0.72693279, 0.72687906, 0.72706173, 0.74231988,
       0.74274969, 0.74298608, 0.74413582, 0.74435072, 0.74531779,
       0.74442594, 0.74485575, 0.74507065, 0.74383495, 0.74430774,
       0.7447268 , 0.74447966, 0.74445817, 0.74413582, 0.74454414,
       0.74508139, 0.74464084, 0.74461935, 0.74414656, 0.74442594,
       0.74461935, 0.74414656, 0.74442594, 0.74368452, 0.74397464,
       0.74385644, 0.73575458, 0.73650674, 0.73667867, 0.74036426,
       0.74130984, 0.74112717, 0.74273895, 0.74393166, 0.74428625,
       0.74308279, 0.74377048, 0.74414656, 0.74386719, 0.74397464,
       0.74480202, 0.74418954, 0.74473755, 0.74471606, 0.74314726,
       0.74371676, 0.74435072, 0.74314726, 0.74371676, 0.74435072,
       0.74356632, 0.74381346, 0.74382421, 0.7261269 , 0.72637404,
       0.7262451 , 0.72540697, 0.72592274, 0.72630957, 0.72733036,
       0.72737334, 0.72719067, 0.72656745, 0.72627733, 0.72647075,
       0.72704024, 0.72697577, 0.72649224, 0.7260087 , 0.72614839,
       0.72684683, 0.72616988, 0.72552517, 0.72538548, 0.72616988,
       0.72552517, 0.72538548, 0.7256756 , 0.72601945, 0.72556815,
       0.74424327, 0.74490947, 0.74531779, 0.74408209, 0.74478053,
       0.74447966, 0.74449041, 0.74510289, 0.74497394, 0.74433998,
       0.74467308, 0.7445119 , 0.74428625, 0.74482351, 0.74498469,
       0.74398539, 0.74474829, 0.74467308, 0.74388868, 0.74435072,
       0.74440445, 0.74388868, 0.74435072, 0.74440445, 0.74342664,
       0.74378123, 0.74403911, 0.73806479, 0.73862354, 0.73877398,
       0.7412776 , 0.74209424, 0.74200827, 0.74402837, 0.74481277,
       0.74499543, 0.74411433, 0.74511363, 0.74488798, 0.74398539,
       0.74455488, 0.74468382, 0.74450116, 0.74474829, 0.74525332,
       0.74395315, 0.74415731, 0.7443937 , 0.74395315, 0.74415731,
       0.7443937 , 0.74433998, 0.7443937 , 0.7445119 ]), 'split1_test_score': array([0.72834823, 0.72874581, 0.72863836, 0.72729519, 0.72724147,
       0.72802587, 0.72823004, 0.72874581, 0.7290037 , 0.72934755,
       0.72885326, 0.72855239, 0.72758532, 0.72780022, 0.72826227,
       0.72752085, 0.72852016, 0.72836972, 0.72832674, 0.72823004,
       0.72824078, 0.72832674, 0.72823004, 0.72824078, 0.72804737,
       0.72838047, 0.72841271, 0.74431574, 0.74504642, 0.74479928,
       0.74401487, 0.744853  , 0.74535803, 0.74453065, 0.74603499,
       0.74528282, 0.74391816, 0.74492822, 0.74474555, 0.74422978,
       0.74468108, 0.74473481, 0.74359581, 0.74430499, 0.74448766,
       0.74249979, 0.74331643, 0.74355282, 0.74249979, 0.74331643,
       0.74355282, 0.74325196, 0.74369251, 0.74411158, 0.74076979,
       0.7416724 , 0.74202699, 0.74334866, 0.74388593, 0.74426201,
       0.7442835 , 0.74559443, 0.745444  , 0.74402562, 0.74510015,
       0.744853  , 0.74323047, 0.74459512, 0.74487449, 0.74501418,
       0.74571263, 0.74521834, 0.74384295, 0.7440686 , 0.74416531,
       0.74384295, 0.7440686 , 0.74416531, 0.74306929, 0.74470257,
       0.74482077, 0.72860612, 0.72799364, 0.7278432 , 0.72753159,
       0.72789693, 0.72811184, 0.72833749, 0.7286491 , 0.72821929,
       0.72793991, 0.72769277, 0.72778948, 0.72668271, 0.72713402,
       0.72753159, 0.72789693, 0.72787544, 0.72793991, 0.72858463,
       0.72785395, 0.72780022, 0.72858463, 0.72785395, 0.72780022,
       0.72737041, 0.72679017, 0.72747786, 0.74384295, 0.74472406,
       0.74511089, 0.74436947, 0.7446381 , 0.74519685, 0.74517536,
       0.74594902, 0.74578784, 0.74397189, 0.74448766, 0.74523983,
       0.74494971, 0.74573412, 0.74548698, 0.74362804, 0.74488524,
       0.74460586, 0.7438322 , 0.74389667, 0.74416531, 0.7438322 ,
       0.74389667, 0.74416531, 0.74360655, 0.74425127, 0.74405785,
       0.74127482, 0.74256426, 0.74258575, 0.7431445 , 0.74410083,
       0.74464884, 0.74384295, 0.74533654, 0.74517536, 0.7441868 ,
       0.74459512, 0.74483151, 0.74435872, 0.74518611, 0.74563741,
       0.74486375, 0.74585232, 0.74543325, 0.74403636, 0.74450915,
       0.74443394, 0.74403636, 0.74450915, 0.74443394, 0.7445199 ,
       0.74460586, 0.74490673, 0.72608098, 0.72729519, 0.72764979,
       0.72743488, 0.72863836, 0.72881028, 0.72886401, 0.72873506,
       0.72850941, 0.72886401, 0.72862761, 0.72826227, 0.72925084,
       0.72894997, 0.72891773, 0.72894997, 0.72840196, 0.72829451,
       0.72858463, 0.72846643, 0.72849867, 0.72858463, 0.72846643,
       0.72849867, 0.72825153, 0.72869208, 0.72885326, 0.743714  ,
       0.74375698, 0.74407934, 0.74465959, 0.74498195, 0.74503567,
       0.74559443, 0.74612095, 0.74624989, 0.74598126, 0.74555145,
       0.74552996, 0.74493897, 0.74592753, 0.74593828, 0.74523983,
       0.74586306, 0.74580934, 0.74445543, 0.74516462, 0.74559443,
       0.74445543, 0.74516462, 0.74559443, 0.74488524, 0.74519685,
       0.74488524, 0.73696596, 0.73761068, 0.73795453, 0.74156494,
       0.74154345, 0.74241382, 0.74421903, 0.74474555, 0.74457363,
       0.74416531, 0.74489599, 0.7446381 , 0.74414382, 0.7445199 ,
       0.74494971, 0.74516462, 0.74575561, 0.74588455, 0.74534729,
       0.74569114, 0.74583083, 0.74534729, 0.74569114, 0.74583083,
       0.74484226, 0.74535803, 0.745444  , 0.72806886, 0.72780022,
       0.72818705, 0.72793991, 0.7276283 , 0.72811184, 0.72763904,
       0.7278432 , 0.72766053, 0.72813333, 0.72744563, 0.72778948,
       0.72767128, 0.7281978 , 0.72813333, 0.72754234, 0.72757457,
       0.72785395, 0.72988481, 0.72869208, 0.728316  , 0.72988481,
       0.72869208, 0.728316  , 0.72874581, 0.72789693, 0.72797215,
       0.74358506, 0.74496046, 0.74512164, 0.74533654, 0.74557294,
       0.74537952, 0.74579859, 0.74579859, 0.74591679, 0.74457363,
       0.74549772, 0.74591679, 0.74462735, 0.7453258 , 0.74561592,
       0.74570188, 0.74629287, 0.74629287, 0.74533654, 0.74568039,
       0.74575561, 0.74533654, 0.74568039, 0.74575561, 0.74468108,
       0.74545474, 0.74543325, 0.73828763, 0.73912576, 0.73907204,
       0.74277916, 0.74275767, 0.74270395, 0.7442835 , 0.74377848,
       0.74460586, 0.74443394, 0.74519685, 0.74522909, 0.74436947,
       0.74551921, 0.7453258 , 0.74586306, 0.74607797, 0.74643256,
       0.7449712 , 0.74572337, 0.74557294, 0.7449712 , 0.74572337,
       0.74557294, 0.745444  , 0.74559443, 0.74573412]), 'split2_test_score': array([0.72668271, 0.72715551, 0.72720923, 0.72636035, 0.72686538,
       0.72681166, 0.72619917, 0.72626365, 0.72685464, 0.72698358,
       0.72719849, 0.72766053, 0.72744563, 0.7273919 , 0.72738116,
       0.72659675, 0.72719849, 0.72748861, 0.7258016 , 0.72651079,
       0.72679017, 0.7258016 , 0.72651079, 0.72679017, 0.72710178,
       0.72752085, 0.72766053, 0.74118886, 0.74203774, 0.74203774,
       0.74144675, 0.74197327, 0.74240308, 0.74196252, 0.74196252,
       0.74183358, 0.74169389, 0.7417906 , 0.74175836, 0.74080203,
       0.74184432, 0.74175836, 0.74109215, 0.74229562, 0.74242457,
       0.74152196, 0.74102768, 0.7413178 , 0.74152196, 0.74102768,
       0.7413178 , 0.74059787, 0.74069458, 0.74093097, 0.73951259,
       0.73973824, 0.74002837, 0.74162942, 0.74274693, 0.74276842,
       0.74300481, 0.74363879, 0.74331643, 0.74199476, 0.74290811,
       0.74309078, 0.74197327, 0.74312301, 0.74294034, 0.74218817,
       0.74247829, 0.74263947, 0.74014657, 0.74114588, 0.74135004,
       0.74014657, 0.74114588, 0.74135004, 0.74057638, 0.74104917,
       0.74135004, 0.72622067, 0.72706954, 0.72713402, 0.72711252,
       0.72622067, 0.72644632, 0.72565116, 0.72559744, 0.72604874,
       0.7261347 , 0.72628514, 0.72614545, 0.72550073, 0.72574787,
       0.72583383, 0.72620992, 0.72688687, 0.72658601, 0.72617768,
       0.72662899, 0.72651079, 0.72617768, 0.72662899, 0.72651079,
       0.72653228, 0.72639259, 0.72662899, 0.74183358, 0.74253202,
       0.74271469, 0.74209146, 0.7430263 , 0.74255351, 0.74217743,
       0.74261798, 0.74246755, 0.74214519, 0.74239233, 0.74254277,
       0.74223115, 0.74255351, 0.74259649, 0.74147898, 0.74197327,
       0.74220966, 0.74125333, 0.74177985, 0.74148973, 0.74125333,
       0.74177985, 0.74148973, 0.74108141, 0.74082352, 0.74084501,
       0.74010358, 0.74084501, 0.74129631, 0.74223115, 0.74325196,
       0.74332717, 0.74364953, 0.74374624, 0.74356357, 0.7430263 ,
       0.74350984, 0.74398264, 0.7432627 , 0.74298332, 0.74356357,
       0.7430263 , 0.74364953, 0.74387518, 0.74174761, 0.74191954,
       0.74208072, 0.74174761, 0.74191954, 0.74208072, 0.74230637,
       0.74227413, 0.74220966, 0.72655377, 0.7272737 , 0.72686538,
       0.72675793, 0.72706954, 0.72701582, 0.72708029, 0.7276283 ,
       0.72713402, 0.7267042 , 0.72675793, 0.72689762, 0.72579085,
       0.72654302, 0.72655377, 0.72574787, 0.72754234, 0.72738116,
       0.72723072, 0.7270588 , 0.72704805, 0.72723072, 0.7270588 ,
       0.72704805, 0.72770352, 0.72715551, 0.72766053, 0.74248904,
       0.74267171, 0.74262873, 0.74297258, 0.7433809 , 0.74327345,
       0.74242457, 0.74346686, 0.74345612, 0.74260724, 0.74330568,
       0.74334866, 0.74299407, 0.74374624, 0.74347761, 0.74374624,
       0.74332717, 0.74324121, 0.74214519, 0.74261798, 0.74228488,
       0.74214519, 0.74261798, 0.74228488, 0.74219892, 0.742575  ,
       0.7428114 , 0.73552609, 0.73620304, 0.73611708, 0.74008209,
       0.74052265, 0.74054414, 0.74209146, 0.74316599, 0.74329494,
       0.74259649, 0.74312301, 0.74319823, 0.74255351, 0.74299407,
       0.74300481, 0.74401487, 0.743714  , 0.74401487, 0.74195178,
       0.7429296 , 0.743714  , 0.74195178, 0.7429296 , 0.743714  ,
       0.74268246, 0.74343463, 0.74327345, 0.72652153, 0.72679017,
       0.72702656, 0.72617768, 0.72666122, 0.72674718, 0.72657526,
       0.72641408, 0.72638184, 0.72585533, 0.72581234, 0.72607023,
       0.72620992, 0.72683315, 0.72683315, 0.72586607, 0.72663973,
       0.72672569, 0.72456589, 0.72653228, 0.72657526, 0.72456589,
       0.72653228, 0.72657526, 0.72560818, 0.72596278, 0.72630663,
       0.74201625, 0.74261798, 0.74280065, 0.74347761, 0.74378922,
       0.74391816, 0.74321972, 0.74374624, 0.74353133, 0.74392891,
       0.74392891, 0.74364953, 0.74306929, 0.74397189, 0.74355282,
       0.74263947, 0.74343463, 0.74368177, 0.74294034, 0.74270395,
       0.74315525, 0.74294034, 0.74270395, 0.74315525, 0.74215594,
       0.74266096, 0.74282214, 0.73719161, 0.73719161, 0.73727757,
       0.74097395, 0.74206997, 0.74204848, 0.74327345, 0.74345612,
       0.74373549, 0.74272544, 0.74309078, 0.74320897, 0.74283289,
       0.74341314, 0.74367102, 0.74385369, 0.74425127, 0.74435872,
       0.74270395, 0.74377848, 0.74404711, 0.74270395, 0.74377848,
       0.74404711, 0.74345612, 0.74420829, 0.74411158]), 'mean_test_score': array([0.72686994, 0.72742511, 0.72751108, 0.72706694, 0.72728901,
       0.72746451, 0.72721737, 0.72752899, 0.72764718, 0.72777971,
       0.72770449, 0.72787642, 0.72771524, 0.7276436 , 0.72769375,
       0.72683054, 0.72777613, 0.72754331, 0.72696665, 0.72727826,
       0.72745735, 0.72696665, 0.72727826, 0.72745735, 0.72753973,
       0.72769375, 0.72770807, 0.74268338, 0.74344629, 0.74343913,
       0.74299857, 0.74377939, 0.74406593, 0.74302722, 0.74379372,
       0.74380088, 0.74308095, 0.7434248 , 0.74340331, 0.74272278,
       0.743346  , 0.74324571, 0.74245056, 0.743346  , 0.74345704,
       0.74197061, 0.74235386, 0.7424828 , 0.74197061, 0.74235386,
       0.7424828 , 0.74233237, 0.74240042, 0.74251862, 0.73992901,
       0.74082803, 0.74103219, 0.74265114, 0.7433066 , 0.74349643,
       0.74370418, 0.74442411, 0.74415906, 0.74328153, 0.74409101,
       0.74400146, 0.74270487, 0.74368627, 0.74399788, 0.74365761,
       0.74413757, 0.74406235, 0.74209597, 0.74274427, 0.74286963,
       0.74209597, 0.74274427, 0.74286963, 0.74186674, 0.74287321,
       0.74311319, 0.72692725, 0.72697381, 0.72702396, 0.72697023,
       0.72695949, 0.72721021, 0.72697381, 0.7270347 , 0.72721737,
       0.72671235, 0.726938  , 0.72700605, 0.72647237, 0.72652609,
       0.72661922, 0.72707768, 0.72721021, 0.72694874, 0.72700605,
       0.7270347 , 0.72695591, 0.72700605, 0.7270347 , 0.72695591,
       0.72706694, 0.72680189, 0.72709559, 0.7429341 , 0.74357523,
       0.74401221, 0.74332093, 0.74404444, 0.74388326, 0.74365403,
       0.74419846, 0.74420562, 0.74315617, 0.74370418, 0.74393699,
       0.7434642 , 0.74401937, 0.74401579, 0.74262965, 0.74346062,
       0.74351792, 0.74255802, 0.74299141, 0.74286963, 0.74255802,
       0.74299141, 0.74286963, 0.74233953, 0.742601  , 0.74237176,
       0.74076714, 0.74174854, 0.74178436, 0.74294126, 0.74371492,
       0.74402295, 0.74356091, 0.74452798, 0.74442411, 0.74373283,
       0.74424502, 0.74445634, 0.74374358, 0.74412682, 0.74457812,
       0.74408742, 0.74459603, 0.74465692, 0.74315975, 0.74333167,
       0.7433854 , 0.74315975, 0.74333167, 0.7433854 , 0.74306662,
       0.74322064, 0.74347136, 0.72608196, 0.72694874, 0.72697381,
       0.72695949, 0.72775464, 0.72788   , 0.72745377, 0.72786209,
       0.72783702, 0.72779045, 0.72776538, 0.72745377, 0.72749675,
       0.72778687, 0.72773315, 0.72745019, 0.72777613, 0.72763644,
       0.72707052, 0.72733199, 0.72745377, 0.72707052, 0.72733199,
       0.72745377, 0.72762927, 0.72757555, 0.72785851, 0.74284097,
       0.74305946, 0.74323138, 0.74392266, 0.74423786, 0.74454231,
       0.74414831, 0.74481452, 0.74492555, 0.74414115, 0.74438829,
       0.74453514, 0.74413757, 0.74471065, 0.74451723, 0.74451007,
       0.74475721, 0.7445638 , 0.74373999, 0.74397639, 0.74410175,
       0.74373999, 0.74397639, 0.74410175, 0.74358956, 0.7439155 ,
       0.74385103, 0.73608221, 0.73677349, 0.73691676, 0.74067043,
       0.74112531, 0.74136171, 0.74301648, 0.74394774, 0.74405161,
       0.74328153, 0.74392983, 0.7439943 , 0.74352151, 0.74382954,
       0.74425218, 0.74445634, 0.74473572, 0.74487183, 0.74348211,
       0.7441125 , 0.74463185, 0.74348211, 0.7441125 , 0.74463185,
       0.74369701, 0.74420204, 0.74418055, 0.72690576, 0.72698814,
       0.7271529 , 0.72650819, 0.72673742, 0.72705619, 0.72718156,
       0.72721021, 0.72707768, 0.72685203, 0.72651177, 0.72677682,
       0.72697381, 0.72733557, 0.7271529 , 0.72647237, 0.72678756,
       0.72714216, 0.72687352, 0.72691651, 0.72675891, 0.72687352,
       0.72691651, 0.72675891, 0.72667653, 0.72662638, 0.72661564,
       0.74328153, 0.74416264, 0.74441336, 0.74429875, 0.74471423,
       0.74459245, 0.74450291, 0.74488257, 0.74480736, 0.74428084,
       0.7446999 , 0.74469274, 0.7439943 , 0.74470707, 0.74471781,
       0.74410891, 0.74482526, 0.74488257, 0.74405519, 0.74424502,
       0.74443844, 0.74405519, 0.74424502, 0.74443844, 0.74342122,
       0.74396564, 0.74409817, 0.73784801, 0.73831364, 0.73837453,
       0.7416769 , 0.74230729, 0.74225357, 0.74386177, 0.74401579,
       0.7444456 , 0.7437579 , 0.74446709, 0.74444202, 0.74372925,
       0.74449574, 0.74456021, 0.7447393 , 0.74502584, 0.7453482 ,
       0.7438761 , 0.74455305, 0.74467125, 0.7438761 , 0.74455305,
       0.74467125, 0.74441336, 0.74473214, 0.74478586]), 'std_test_score': array([0.0011383 , 0.00098686, 0.00082527, 0.00050995, 0.00036684,
       0.00049991, 0.0008291 , 0.00101392, 0.00096373, 0.00110867,
       0.00081422, 0.00048827, 0.0002882 , 0.00017974, 0.00040267,
       0.00049651, 0.00055225, 0.00065357, 0.00104007, 0.00071387,
       0.00059789, 0.00104007, 0.00071387, 0.00059789, 0.00038917,
       0.00050514, 0.00055694, 0.00128026, 0.00123573, 0.00112777,
       0.00111472, 0.00128469, 0.00123451, 0.00109336, 0.00168767,
       0.00144937, 0.00098779, 0.00128426, 0.00123816, 0.00142963,
       0.00116408, 0.00121513, 0.00103317, 0.00082286, 0.00084226,
       0.00040323, 0.0009691 , 0.00091491, 0.00040323, 0.0009691 ,
       0.00091491, 0.00122722, 0.00125838, 0.00129848, 0.00059453,
       0.00080846, 0.00081596, 0.00073837, 0.0004652 , 0.00061034,
       0.00052888, 0.00084352, 0.00092317, 0.00091359, 0.00090329,
       0.00072064, 0.00053352, 0.00064877, 0.00079991, 0.00115647,
       0.00132175, 0.0010696 , 0.00151581, 0.00120883, 0.00116018,
       0.00151581, 0.00120883, 0.00116018, 0.00101962, 0.0014915 ,
       0.00141746, 0.00119208, 0.00087439, 0.00071807, 0.00052615,
       0.00069859, 0.00068688, 0.00109707, 0.00125211, 0.00089393,
       0.00086852, 0.00057916, 0.00067339, 0.00072293, 0.00057856,
       0.0006989 , 0.00068957, 0.00047045, 0.00070921, 0.00111668,
       0.0005793 , 0.00059732, 0.00111668, 0.0005793 , 0.00059732,
       0.00037921, 0.00033908, 0.00035165, 0.00083144, 0.00089801,
       0.00098829, 0.00093875, 0.00072324, 0.0010792 , 0.00122431,
       0.0013652 , 0.00135997, 0.00075842, 0.00093344, 0.00110297,
       0.00112411, 0.00131035, 0.0011806 , 0.00088393, 0.00118963,
       0.00099057, 0.00105304, 0.00089091, 0.00109392, 0.00105304,
       0.00089091, 0.00109392, 0.0010309 , 0.00140226, 0.00131647,
       0.00049069, 0.00070462, 0.00057114, 0.0005172 , 0.00035081,
       0.00054183, 0.00027374, 0.00064951, 0.00066253, 0.00050631,
       0.00052004, 0.00035349, 0.00045741, 0.00090125, 0.00084723,
       0.00077669, 0.00092558, 0.00063609, 0.00100819, 0.00107015,
       0.00097762, 0.00100819, 0.00107015, 0.00097762, 0.00102799,
       0.0010012 , 0.0011079 , 0.00038483, 0.00047484, 0.00051342,
       0.00033745, 0.00065568, 0.00073408, 0.0010333 , 0.00063909,
       0.00056192, 0.00088178, 0.00077017, 0.00058499, 0.00141294,
       0.00098429, 0.00096509, 0.00131507, 0.00044722, 0.0004692 ,
       0.00130658, 0.00083732, 0.00074498, 0.00130658, 0.00083732,
       0.00074498, 0.00054093, 0.00079753, 0.00074467, 0.00062117,
       0.00049425, 0.00061709, 0.00070502, 0.00065848, 0.00090458,
       0.00130889, 0.00108392, 0.00114516, 0.00139435, 0.0009186 ,
       0.00090076, 0.00083003, 0.00090823, 0.00104014, 0.00061023,
       0.00106035, 0.00104985, 0.00112968, 0.0010466 , 0.00137042,
       0.00112968, 0.0010466 , 0.00137042, 0.00109874, 0.00107118,
       0.00084665, 0.00063183, 0.00060483, 0.00076879, 0.00064292,
       0.00043669, 0.0007811 , 0.00089047, 0.00064495, 0.00054776,
       0.0006557 , 0.00073253, 0.0005976 , 0.00069373, 0.00063131,
       0.00088408, 0.00050588, 0.00083348, 0.0007712 , 0.00140629,
       0.0011616 , 0.00088676, 0.00140629, 0.0011616 , 0.00088676,
       0.00088656, 0.00083191, 0.00092125, 0.00083806, 0.00059883,
       0.00079782, 0.00106015, 0.00069837, 0.00076753, 0.00044685,
       0.00059473, 0.0005281 , 0.00095151, 0.00068708, 0.00073449,
       0.00059844, 0.00061246, 0.0007071 , 0.00075882, 0.00059155,
       0.00050574, 0.00222771, 0.00132113, 0.00120341, 0.00222771,
       0.00132113, 0.00120341, 0.00146346, 0.00089871, 0.00100546,
       0.00093417, 0.00109243, 0.00114316, 0.00077421, 0.00072971,
       0.0006019 , 0.00105285, 0.00085223, 0.00098095, 0.00026651,
       0.00064074, 0.00093439, 0.00066874, 0.00055883, 0.00086314,
       0.00125327, 0.00116814, 0.00107622, 0.0009853 , 0.00121742,
       0.00106186, 0.0009853 , 0.00121742, 0.00106186, 0.00103089,
       0.00114798, 0.0010668 , 0.00047298, 0.00081946, 0.00078515,
       0.00078921, 0.00031862, 0.00031889, 0.00042885, 0.00057871,
       0.0005267 , 0.00074163, 0.00097379, 0.00088294, 0.00065293,
       0.00086082, 0.00068119, 0.00083743, 0.00077114, 0.0008493 ,
       0.0009272 , 0.00084187, 0.0006531 , 0.0009272 , 0.00084187,
       0.0006531 , 0.0008132 , 0.00061441, 0.00069014]), 'rank_test_score': array([305, 253, 243, 274, 257, 245, 260, 242, 235, 224, 232, 218, 230,
       236, 233, 307, 225, 240, 290, 258, 246, 290, 258, 246, 241, 233,
       231, 178, 135, 136, 163, 109,  78, 161, 108, 107, 158, 137, 139,
       176, 142, 151, 187, 142, 134, 198, 190, 185, 198, 190, 185, 193,
       188, 184, 210, 207, 206, 179, 147, 128, 117,  47,  64, 148,  76,
        89, 177, 120,  90, 121,  67,  79, 196, 174, 169, 196, 174, 169,
       200, 168, 157, 299, 285, 280, 289, 292, 262, 285, 277, 260, 314,
       298, 281, 322, 319, 317, 270, 262, 296, 281, 277, 294, 281, 277,
       294, 274, 308, 269, 167, 124,  88, 146,  83, 101, 122,  61,  59,
       156, 117,  97, 132,  85,  86, 180, 133, 127, 182, 164, 169, 182,
       164, 169, 192, 181, 189, 208, 202, 201, 166, 116,  84, 125,  35,
        47, 114,  55,  41, 111,  69,  28,  77,  26,  23, 154, 144, 140,
       154, 144, 140, 159, 153, 131, 324, 296, 285, 292, 228, 217, 248,
       219, 221, 222, 227, 248, 244, 223, 229, 252, 225, 237, 272, 255,
       248, 272, 255, 248, 238, 239, 220, 173, 160, 152,  99,  58,  33,
        65,   8,   3,  66,  51,  34,  67,  17,  36,  37,  11,  29, 112,
        93,  73, 112,  93,  73, 123, 100, 105, 216, 215, 214, 209, 205,
       204, 162,  96,  82, 148,  98,  91, 126, 106,  54,  41,  13,   6,
       129,  70,  24, 129,  70,  24, 119,  60,  62, 302, 284, 266, 321,
       313, 276, 265, 262, 270, 306, 320, 310, 285, 254, 266, 322, 309,
       268, 303, 300, 311, 303, 300, 311, 315, 316, 318, 148,  63,  49,
        52,  16,  27,  38,   4,   9,  53,  19,  20,  91,  18,  15,  72,
         7,   4,  80,  55,  45,  80,  55,  45, 138,  95,  75, 213, 212,
       211, 203, 194, 195, 104,  86,  43, 110,  40,  44, 115,  39,  30,
        12,   2,   1, 102,  31,  21, 102,  31,  21,  49,  14,  10])}

del sub_train_set
train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')

params = grid.best_params_
params['n_jobs'] = -1
params['verbose'] = 5
best_model = _RandomForestClassifier(**params)
best_model.get_params()

{'bootstrap': False,
 'class_weight': None,
 'criterion': 'entropy',
 'max_depth': 50,
 'max_features': 'auto',
 'max_leaf_nodes': None,
 'min_impurity_decrease': 0.0,
 'min_impurity_split': None,
 'min_samples_leaf': 2,
 'min_samples_split': 10,
 'min_weight_fraction_leaf': 0.0,
 'n_estimators': 1000,
 'n_jobs': -1,
 'oob_score': False,
 'random_state': None,
 'verbose': 5,
 'warm_start': False}

best_model.fit(train_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])

[Parallel(n_jobs=-1)]: Using backend ThreadingBackend with 12 concurrent workers.

building tree 1 of 1000
building tree 2 of 1000
building tree 3 of 1000
building tree 4 of 1000
building tree 5 of 1000
building tree 6 of 1000
building tree 7 of 1000
building tree 8 of 1000
building tree 9 of 1000
building tree 10 of 1000
building tree 11 of 1000
building tree 12 of 1000
building tree 13 of 1000
building tree 14 of 1000
building tree 15 of 1000
building tree 16 of 1000
building tree 17 of 1000
building tree 18 of 1000
building tree 19 of 1000
building tree 20 of 1000
building tree 21 of 1000
building tree 22 of 1000
building tree 23 of 1000
building tree 24 of 1000
building tree 25 of 1000
building tree 26 of 1000
building tree 27 of 1000
building tree 28 of 1000
building tree 29 of 1000
building tree 30 of 1000
building tree 31 of 1000
building tree 32 of 1000
building tree 33 of 1000
building tree 34 of 1000
building tree 35 of 1000
building tree 36 of 1000
building tree 37 of 1000
building tree 38 of 1000
building tree 39 of 1000
building tree 40 of 1000
building tree 41 of 1000
building tree 42 of 1000
building tree 43 of 1000
building tree 44 of 1000
building tree 45 of 1000
building tree 46 of 1000
building tree 47 of 1000
building tree 48 of 1000
building tree 49 of 1000
building tree 50 of 1000
building tree 51 of 1000
building tree 52 of 1000
building tree 53 of 1000
building tree 54 of 1000
building tree 55 of 1000
building tree 56 of 1000
building tree 57 of 1000
building tree 58 of 1000
building tree 59 of 1000
building tree 60 of 1000

[Parallel(n_jobs=-1)]: Done  48 tasks      | elapsed:  2.0min

building tree 61 of 1000
building tree 62 of 1000
building tree 63 of 1000
building tree 64 of 1000
building tree 65 of 1000
building tree 66 of 1000
building tree 67 of 1000
building tree 68 of 1000
building tree 69 of 1000
building tree 70 of 1000
building tree 71 of 1000
building tree 72 of 1000
building tree 73 of 1000
building tree 74 of 1000
building tree 75 of 1000
building tree 76 of 1000
building tree 77 of 1000
building tree 78 of 1000
building tree 79 of 1000
building tree 80 of 1000
building tree 81 of 1000
building tree 82 of 1000
building tree 83 of 1000
building tree 84 of 1000
building tree 85 of 1000
building tree 86 of 1000
building tree 87 of 1000
building tree 88 of 1000
building tree 89 of 1000
building tree 90 of 1000
building tree 91 of 1000
building tree 92 of 1000
building tree 93 of 1000
building tree 94 of 1000
building tree 95 of 1000
building tree 96 of 1000
building tree 97 of 1000
building tree 98 of 1000
building tree 99 of 1000
building tree 100 of 1000
building tree 101 of 1000
building tree 102 of 1000
building tree 103 of 1000
building tree 104 of 1000
building tree 105 of 1000
building tree 106 of 1000
building tree 107 of 1000
building tree 108 of 1000
building tree 109 of 1000
building tree 110 of 1000
building tree 111 of 1000
building tree 112 of 1000
building tree 113 of 1000
building tree 114 of 1000
building tree 115 of 1000
building tree 116 of 1000
building tree 117 of 1000
building tree 118 of 1000
building tree 119 of 1000
building tree 120 of 1000
building tree 121 of 1000
building tree 122 of 1000
building tree 123 of 1000
building tree 124 of 1000
building tree 125 of 1000
building tree 126 of 1000
building tree 127 of 1000
building tree 128 of 1000
building tree 129 of 1000
building tree 130 of 1000
building tree 131 of 1000
building tree 132 of 1000
building tree 133 of 1000
building tree 134 of 1000
building tree 135 of 1000
building tree 136 of 1000
building tree 137 of 1000
building tree 138 of 1000
building tree 139 of 1000
building tree 140 of 1000
building tree 141 of 1000
building tree 142 of 1000
building tree 143 of 1000
building tree 144 of 1000
building tree 145 of 1000
building tree 146 of 1000
building tree 147 of 1000
building tree 148 of 1000
building tree 149 of 1000
building tree 150 of 1000

[Parallel(n_jobs=-1)]: Done 138 tasks      | elapsed:  5.6min

building tree 151 of 1000
building tree 152 of 1000
building tree 153 of 1000
building tree 154 of 1000
building tree 155 of 1000
building tree 156 of 1000
building tree 157 of 1000
building tree 158 of 1000
building tree 159 of 1000
building tree 160 of 1000
building tree 161 of 1000
building tree 162 of 1000
building tree 163 of 1000
building tree 164 of 1000
building tree 165 of 1000
building tree 166 of 1000
building tree 167 of 1000
building tree 168 of 1000
building tree 169 of 1000
building tree 170 of 1000
building tree 171 of 1000
building tree 172 of 1000
building tree 173 of 1000
building tree 174 of 1000
building tree 175 of 1000
building tree 176 of 1000
building tree 177 of 1000
building tree 178 of 1000
building tree 179 of 1000
building tree 180 of 1000
building tree 181 of 1000
building tree 182 of 1000
building tree 183 of 1000
building tree 184 of 1000
building tree 185 of 1000
building tree 186 of 1000
building tree 187 of 1000
building tree 188 of 1000
building tree 189 of 1000
building tree 190 of 1000
building tree 191 of 1000
building tree 192 of 1000
building tree 193 of 1000
building tree 194 of 1000
building tree 195 of 1000
building tree 196 of 1000
building tree 197 of 1000
building tree 198 of 1000
building tree 199 of 1000
building tree 200 of 1000
building tree 201 of 1000
building tree 202 of 1000
building tree 203 of 1000
building tree 204 of 1000
building tree 205 of 1000
building tree 206 of 1000
building tree 207 of 1000
building tree 208 of 1000
building tree 209 of 1000
building tree 210 of 1000
building tree 211 of 1000
building tree 212 of 1000
building tree 213 of 1000
building tree 214 of 1000
building tree 215 of 1000
building tree 216 of 1000
building tree 217 of 1000
building tree 218 of 1000
building tree 219 of 1000
building tree 220 of 1000
building tree 221 of 1000
building tree 222 of 1000
building tree 223 of 1000
building tree 224 of 1000
building tree 225 of 1000
building tree 226 of 1000
building tree 227 of 1000
building tree 228 of 1000
building tree 229 of 1000
building tree 230 of 1000
building tree 231 of 1000
building tree 232 of 1000
building tree 233 of 1000
building tree 234 of 1000
building tree 235 of 1000
building tree 236 of 1000
building tree 237 of 1000
building tree 238 of 1000
building tree 239 of 1000
building tree 240 of 1000
building tree 241 of 1000
building tree 242 of 1000
building tree 243 of 1000
building tree 244 of 1000
building tree 245 of 1000
building tree 246 of 1000
building tree 247 of 1000
building tree 248 of 1000
building tree 249 of 1000
building tree 250 of 1000
building tree 251 of 1000
building tree 252 of 1000
building tree 253 of 1000
building tree 254 of 1000
building tree 255 of 1000
building tree 256 of 1000
building tree 257 of 1000
building tree 258 of 1000
building tree 259 of 1000
building tree 260 of 1000
building tree 261 of 1000
building tree 262 of 1000
building tree 263 of 1000
building tree 264 of 1000
building tree 265 of 1000
building tree 266 of 1000
building tree 267 of 1000
building tree 268 of 1000
building tree 269 of 1000
building tree 270 of 1000
building tree 271 of 1000
building tree 272 of 1000
building tree 273 of 1000
building tree 274 of 1000
building tree 275 of 1000

[Parallel(n_jobs=-1)]: Done 264 tasks      | elapsed: 10.3min

building tree 276 of 1000
building tree 277 of 1000
building tree 278 of 1000
building tree 279 of 1000
building tree 280 of 1000
building tree 281 of 1000
building tree 282 of 1000
building tree 283 of 1000
building tree 284 of 1000
building tree 285 of 1000
building tree 286 of 1000
building tree 287 of 1000
building tree 288 of 1000
building tree 289 of 1000
building tree 290 of 1000
building tree 291 of 1000
building tree 292 of 1000
building tree 293 of 1000
building tree 294 of 1000
building tree 295 of 1000
building tree 296 of 1000
building tree 297 of 1000
building tree 298 of 1000
building tree 299 of 1000
building tree 300 of 1000
building tree 301 of 1000
building tree 302 of 1000
building tree 303 of 1000
building tree 304 of 1000
building tree 305 of 1000
building tree 306 of 1000
building tree 307 of 1000
building tree 308 of 1000
building tree 309 of 1000
building tree 310 of 1000
building tree 311 of 1000
building tree 312 of 1000
building tree 313 of 1000
building tree 314 of 1000
building tree 315 of 1000
building tree 316 of 1000
building tree 317 of 1000
building tree 318 of 1000
building tree 319 of 1000
building tree 320 of 1000
building tree 321 of 1000
building tree 322 of 1000
building tree 323 of 1000
building tree 324 of 1000
building tree 325 of 1000
building tree 326 of 1000
building tree 327 of 1000
building tree 328 of 1000
building tree 329 of 1000
building tree 330 of 1000
building tree 331 of 1000
building tree 332 of 1000
building tree 333 of 1000
building tree 334 of 1000
building tree 335 of 1000
building tree 336 of 1000
building tree 337 of 1000
building tree 338 of 1000
building tree 339 of 1000
building tree 340 of 1000
building tree 341 of 1000
building tree 342 of 1000
building tree 343 of 1000
building tree 344 of 1000
building tree 345 of 1000
building tree 346 of 1000
building tree 347 of 1000
building tree 348 of 1000
building tree 349 of 1000
building tree 350 of 1000
building tree 351 of 1000
building tree 352 of 1000
building tree 353 of 1000
building tree 354 of 1000
building tree 355 of 1000
building tree 356 of 1000
building tree 357 of 1000
building tree 358 of 1000
building tree 359 of 1000
building tree 360 of 1000
building tree 361 of 1000
building tree 362 of 1000
building tree 363 of 1000
building tree 364 of 1000
building tree 365 of 1000
building tree 366 of 1000
building tree 367 of 1000
building tree 368 of 1000
building tree 369 of 1000
building tree 370 of 1000
building tree 371 of 1000
building tree 372 of 1000
building tree 373 of 1000
building tree 374 of 1000
building tree 375 of 1000
building tree 376 of 1000
building tree 377 of 1000
building tree 378 of 1000
building tree 379 of 1000
building tree 380 of 1000
building tree 381 of 1000
building tree 382 of 1000
building tree 383 of 1000
building tree 384 of 1000
building tree 385 of 1000
building tree 386 of 1000
building tree 387 of 1000
building tree 388 of 1000
building tree 389 of 1000
building tree 390 of 1000
building tree 391 of 1000
building tree 392 of 1000
building tree 393 of 1000
building tree 394 of 1000
building tree 395 of 1000
building tree 396 of 1000
building tree 397 of 1000
building tree 398 of 1000
building tree 399 of 1000
building tree 400 of 1000
building tree 401 of 1000
building tree 402 of 1000
building tree 403 of 1000
building tree 404 of 1000
building tree 405 of 1000
building tree 406 of 1000
building tree 407 of 1000
building tree 408 of 1000
building tree 409 of 1000
building tree 410 of 1000
building tree 411 of 1000
building tree 412 of 1000
building tree 413 of 1000
building tree 414 of 1000
building tree 415 of 1000
building tree 416 of 1000
building tree 417 of 1000
building tree 418 of 1000
building tree 419 of 1000
building tree 420 of 1000
building tree 421 of 1000
building tree 422 of 1000
building tree 423 of 1000
building tree 424 of 1000
building tree 425 of 1000
building tree 426 of 1000
building tree 427 of 1000
building tree 428 of 1000
building tree 429 of 1000
building tree 430 of 1000
building tree 431 of 1000
building tree 432 of 1000
building tree 433 of 1000
building tree 434 of 1000
building tree 435 of 1000
building tree 436 of 1000
building tree 437 of 1000

[Parallel(n_jobs=-1)]: Done 426 tasks      | elapsed: 16.2min

building tree 438 of 1000
building tree 439 of 1000
building tree 440 of 1000
building tree 441 of 1000
building tree 442 of 1000
building tree 443 of 1000
building tree 444 of 1000
building tree 445 of 1000
building tree 446 of 1000
building tree 447 of 1000
building tree 448 of 1000
building tree 449 of 1000
building tree 450 of 1000
building tree 451 of 1000
building tree 452 of 1000
building tree 453 of 1000
building tree 454 of 1000
building tree 455 of 1000
building tree 456 of 1000
building tree 457 of 1000
building tree 458 of 1000
building tree 459 of 1000
building tree 460 of 1000
building tree 461 of 1000
building tree 462 of 1000
building tree 463 of 1000
building tree 464 of 1000
building tree 465 of 1000
building tree 466 of 1000
building tree 467 of 1000
building tree 468 of 1000
building tree 469 of 1000
building tree 470 of 1000
building tree 471 of 1000
building tree 472 of 1000
building tree 473 of 1000
building tree 474 of 1000
building tree 475 of 1000
building tree 476 of 1000
building tree 477 of 1000
building tree 478 of 1000
building tree 479 of 1000
building tree 480 of 1000
building tree 481 of 1000
building tree 482 of 1000
building tree 483 of 1000
building tree 484 of 1000
building tree 485 of 1000
building tree 486 of 1000
building tree 487 of 1000
building tree 488 of 1000
building tree 489 of 1000
building tree 490 of 1000
building tree 491 of 1000
building tree 492 of 1000
building tree 493 of 1000
building tree 494 of 1000
building tree 495 of 1000
building tree 496 of 1000
building tree 497 of 1000
building tree 498 of 1000
building tree 499 of 1000
building tree 500 of 1000
building tree 501 of 1000
building tree 502 of 1000
building tree 503 of 1000
building tree 504 of 1000
building tree 505 of 1000
building tree 506 of 1000
building tree 507 of 1000
building tree 508 of 1000
building tree 509 of 1000
building tree 510 of 1000
building tree 511 of 1000
building tree 512 of 1000
building tree 513 of 1000
building tree 514 of 1000
building tree 515 of 1000
building tree 516 of 1000
building tree 517 of 1000
building tree 518 of 1000
building tree 519 of 1000
building tree 520 of 1000
building tree 521 of 1000
building tree 522 of 1000
building tree 523 of 1000
building tree 524 of 1000
building tree 525 of 1000
building tree 526 of 1000
building tree 527 of 1000
building tree 528 of 1000
building tree 529 of 1000
building tree 530 of 1000
building tree 531 of 1000
building tree 532 of 1000
building tree 533 of 1000
building tree 534 of 1000
building tree 535 of 1000
building tree 536 of 1000
building tree 537 of 1000
building tree 538 of 1000
building tree 539 of 1000
building tree 540 of 1000
building tree 541 of 1000
building tree 542 of 1000
building tree 543 of 1000
building tree 544 of 1000
building tree 545 of 1000
building tree 546 of 1000
building tree 547 of 1000
building tree 548 of 1000
building tree 549 of 1000
building tree 550 of 1000
building tree 551 of 1000
building tree 552 of 1000
building tree 553 of 1000
building tree 554 of 1000
building tree 555 of 1000
building tree 556 of 1000
building tree 557 of 1000
building tree 558 of 1000
building tree 559 of 1000
building tree 560 of 1000
building tree 561 of 1000
building tree 562 of 1000
building tree 563 of 1000
building tree 564 of 1000
building tree 565 of 1000
building tree 566 of 1000
building tree 567 of 1000
building tree 568 of 1000
building tree 569 of 1000
building tree 570 of 1000
building tree 571 of 1000
building tree 572 of 1000
building tree 573 of 1000
building tree 574 of 1000
building tree 575 of 1000
building tree 576 of 1000
building tree 577 of 1000
building tree 578 of 1000
building tree 579 of 1000
building tree 580 of 1000
building tree 581 of 1000
building tree 582 of 1000
building tree 583 of 1000
building tree 584 of 1000
building tree 585 of 1000
building tree 586 of 1000
building tree 587 of 1000
building tree 588 of 1000
building tree 589 of 1000
building tree 590 of 1000
building tree 591 of 1000
building tree 592 of 1000
building tree 593 of 1000
building tree 594 of 1000
building tree 595 of 1000
building tree 596 of 1000
building tree 597 of 1000
building tree 598 of 1000
building tree 599 of 1000
building tree 600 of 1000
building tree 601 of 1000
building tree 602 of 1000
building tree 603 of 1000
building tree 604 of 1000
building tree 605 of 1000
building tree 606 of 1000
building tree 607 of 1000
building tree 608 of 1000
building tree 609 of 1000
building tree 610 of 1000
building tree 611 of 1000
building tree 612 of 1000
building tree 613 of 1000
building tree 614 of 1000
building tree 615 of 1000
building tree 616 of 1000
building tree 617 of 1000
building tree 618 of 1000
building tree 619 of 1000
building tree 620 of 1000
building tree 621 of 1000
building tree 622 of 1000
building tree 623 of 1000
building tree 624 of 1000
building tree 625 of 1000
building tree 626 of 1000
building tree 627 of 1000
building tree 628 of 1000
building tree 629 of 1000
building tree 630 of 1000
building tree 631 of 1000
building tree 632 of 1000
building tree 633 of 1000
building tree 634 of 1000
building tree 635 of 1000

_jl.dump(best_model, "../models/best_Random_Forest_2.joblib")

['../models/best_Random_Forest_2.joblib']

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_3.pickle')
test_set.head()

# test classifier
predic = best_model.predict(test_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
print("predictions:\n", predic)

[Parallel(n_jobs=12)]: Using backend ThreadingBackend with 12 concurrent workers.
[Parallel(n_jobs=12)]: Done  48 tasks      | elapsed:    1.8s
[Parallel(n_jobs=12)]: Done 138 tasks      | elapsed:    4.8s
[Parallel(n_jobs=12)]: Done 264 tasks      | elapsed:    8.9s
[Parallel(n_jobs=12)]: Done 426 tasks      | elapsed:   13.3s
[Parallel(n_jobs=12)]: Done 624 tasks      | elapsed:   18.8s
[Parallel(n_jobs=12)]: Done 858 tasks      | elapsed:   25.2s
[Parallel(n_jobs=12)]: Done 1000 out of 1000 | elapsed:   28.9s finished

predictions:
 [0 0 1 ... 0 1 0]

set(predic)

{0, 1}

# evaluate classifier

print("Report for Random Forest classifier:")
print(_classification_report(test_set['likes'], predic))

print("Accuracy for Random Forest Classifier:", _accuracy_score(test_set['likes'], predic)*100)

Report for Random Forest classifier:
              precision    recall  f1-score   support

           0       0.69      0.40      0.51     50930
           1       0.75      0.91      0.83    103063

    accuracy                           0.74    153993
   macro avg       0.72      0.66      0.67    153993
weighted avg       0.73      0.74      0.72    153993

Accuracy for Random Forest Classifier: 74.16765697142078

# Confusion matrix for Random Forest

print("Confusion Matrix for Random Forest: ")
_confusion_matrix(test_set['likes'], predic)

Confusion Matrix for Random Forest before balance the data:

array([[20372, 30558],
       [ 9222, 93841]], dtype=int64)

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], predic)

_plt.plot(fpr,tpr)
_plt.xlim([0.0,1.0])
_plt.ylim([0.0,1.0])

_plt.title("Random Forest ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

%reset

6.3 Deep Learning¶

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')
train_set.head()

Rule for calculating the number of hidden layers (from here):

$N_h = \frac {N_s} {(\alpha * (N_i + N_o))}$

$N_h$ is the number of hidden neurons
$N_i$ is the number of input neurons
$N_o$ is the number of output neurons
$N_s$ number of samples in training data set
$\alpha$ an arbitrary scaling factor usually between 5 and 10

# Number of neurons for hiddens layer
number_samples, number_features = train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']).shape

number_input_neurons = number_features
number_output_neurons = 1
number_train_semples = number_samples
alpha = 6

number_hidden_neurons = round(number_train_semples / (alpha * (number_input_neurons + number_output_neurons)))
print(number_hidden_neurons)

119

classifier = _Sequential()

# First Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal', input_dim = number_features))

# Second Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Third Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Output Layer
classifier.add(_Dense(1, activation = 'sigmoid', kernel_initializer = 'random_normal'))

# Compiling the neural network
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

# Fitting the data to the training dataset
classifier.fit(train_set.drop(
    columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']),
    train_set['likes'], validation_split = 0.3, batch_size = 100, epochs = 100)

Train on 390870 samples, validate on 167516 samples
Epoch 1/100
390870/390870 [==============================] - 206s 527us/step - loss: 0.6102 - acc: 0.6933 - val_loss: 0.5960 - val_acc: 0.7106
Epoch 2/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5606 - acc: 0.7197 - val_loss: 0.5684 - val_acc: 0.7193
Epoch 3/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5503 - acc: 0.7273 - val_loss: 0.5508 - val_acc: 0.7308
Epoch 4/100
390870/390870 [==============================] - 24s 63us/step - loss: 0.5454 - acc: 0.7317 - val_loss: 0.5547 - val_acc: 0.7287
Epoch 5/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5414 - acc: 0.7337 - val_loss: 0.5375 - val_acc: 0.7373
Epoch 6/100
390870/390870 [==============================] - 25s 63us/step - loss: 0.5393 - acc: 0.7350 - val_loss: 0.5453 - val_acc: 0.7272
Epoch 7/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5381 - acc: 0.7366 - val_loss: 0.5348 - val_acc: 0.7402
Epoch 8/100
390870/390870 [==============================] - 24s 61us/step - loss: 0.5354 - acc: 0.7380 - val_loss: 0.5390 - val_acc: 0.7395
Epoch 9/100
390870/390870 [==============================] - 24s 62us/step - loss: 0.5340 - acc: 0.7397 - val_loss: 0.5313 - val_acc: 0.7390
Epoch 10/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5322 - acc: 0.7405 - val_loss: 0.5354 - val_acc: 0.7448
Epoch 11/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5326 - acc: 0.7408 - val_loss: 0.5372 - val_acc: 0.7415
Epoch 12/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5298 - acc: 0.7423 - val_loss: 0.5361 - val_acc: 0.7384
Epoch 13/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5300 - acc: 0.7420 - val_loss: 0.5407 - val_acc: 0.7356
Epoch 14/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5290 - acc: 0.7425 - val_loss: 0.5250 - val_acc: 0.7481
Epoch 15/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5274 - acc: 0.7433 - val_loss: 0.5267 - val_acc: 0.7436
Epoch 16/100
390870/390870 [==============================] - 24s 62us/step - loss: 0.5255 - acc: 0.7449 - val_loss: 0.5217 - val_acc: 0.7494
Epoch 17/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5254 - acc: 0.7451 - val_loss: 0.5313 - val_acc: 0.7460
Epoch 18/100
390870/390870 [==============================] - 24s 61us/step - loss: 0.5241 - acc: 0.7460 - val_loss: 0.5246 - val_acc: 0.7471
Epoch 19/100
390870/390870 [==============================] - 24s 61us/step - loss: 0.5240 - acc: 0.7460 - val_loss: 0.5237 - val_acc: 0.7465
Epoch 20/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5234 - acc: 0.7461 - val_loss: 0.5250 - val_acc: 0.7481
Epoch 21/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5226 - acc: 0.7462 - val_loss: 0.5417 - val_acc: 0.7359
Epoch 22/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5234 - acc: 0.7465 - val_loss: 0.5258 - val_acc: 0.7474
Epoch 23/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5222 - acc: 0.7472 - val_loss: 0.5217 - val_acc: 0.7499
Epoch 24/100
390870/390870 [==============================] - 24s 61us/step - loss: 0.5210 - acc: 0.7478 - val_loss: 0.5267 - val_acc: 0.7442
Epoch 25/100
390870/390870 [==============================] - 24s 61us/step - loss: 0.5207 - acc: 0.7484 - val_loss: 0.5455 - val_acc: 0.7354
Epoch 26/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5202 - acc: 0.7486 - val_loss: 0.5288 - val_acc: 0.7449
Epoch 27/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5214 - acc: 0.7476 - val_loss: 0.5456 - val_acc: 0.7338
Epoch 28/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5209 - acc: 0.7484 - val_loss: 0.5258 - val_acc: 0.7468
Epoch 29/100
390870/390870 [==============================] - 24s 61us/step - loss: 0.5204 - acc: 0.7472 - val_loss: 0.5231 - val_acc: 0.7481
Epoch 30/100
390870/390870 [==============================] - 24s 61us/step - loss: 0.5190 - acc: 0.7493 - val_loss: 0.5230 - val_acc: 0.7463
Epoch 31/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5189 - acc: 0.7490 - val_loss: 0.5250 - val_acc: 0.7480
Epoch 32/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5205 - acc: 0.7482 - val_loss: 0.5226 - val_acc: 0.7496
Epoch 33/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5202 - acc: 0.7486 - val_loss: 0.5276 - val_acc: 0.7457
Epoch 34/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5172 - acc: 0.7503 - val_loss: 0.5243 - val_acc: 0.7481
Epoch 35/100
390870/390870 [==============================] - 25s 63us/step - loss: 0.5173 - acc: 0.7503 - val_loss: 0.5219 - val_acc: 0.7481
Epoch 36/100
390870/390870 [==============================] - 24s 62us/step - loss: 0.5174 - acc: 0.7494 - val_loss: 0.5212 - val_acc: 0.7479
Epoch 37/100
390870/390870 [==============================] - 24s 62us/step - loss: 0.5172 - acc: 0.7500 - val_loss: 0.5314 - val_acc: 0.7419
Epoch 38/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5190 - acc: 0.7495 - val_loss: 0.5206 - val_acc: 0.7498
Epoch 39/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5170 - acc: 0.7502 - val_loss: 0.5233 - val_acc: 0.7471
Epoch 40/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5176 - acc: 0.7498 - val_loss: 0.5207 - val_acc: 0.7506
Epoch 41/100
390870/390870 [==============================] - 24s 61us/step - loss: 0.5187 - acc: 0.7491 - val_loss: 0.5242 - val_acc: 0.7474
Epoch 42/100
390870/390870 [==============================] - 24s 61us/step - loss: 0.5160 - acc: 0.7508 - val_loss: 0.5182 - val_acc: 0.7502
Epoch 43/100
390870/390870 [==============================] - 24s 62us/step - loss: 0.5154 - acc: 0.7512 - val_loss: 0.5214 - val_acc: 0.7492
Epoch 44/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5152 - acc: 0.7511 - val_loss: 0.5259 - val_acc: 0.7432
Epoch 45/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5155 - acc: 0.7511 - val_loss: 0.5198 - val_acc: 0.7497
Epoch 46/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5165 - acc: 0.7506 - val_loss: 0.5294 - val_acc: 0.7464
Epoch 47/100
390870/390870 [==============================] - 24s 61us/step - loss: 0.5163 - acc: 0.7510 - val_loss: 0.5256 - val_acc: 0.7462
Epoch 48/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5156 - acc: 0.7510 - val_loss: 0.5188 - val_acc: 0.7501
Epoch 49/100
390870/390870 [==============================] - 23s 60us/step - loss: 0.5160 - acc: 0.7499 - val_loss: 0.5257 - val_acc: 0.7473
Epoch 50/100
390870/390870 [==============================] - 24s 61us/step - loss: 0.5167 - acc: 0.7518 - val_loss: 0.5215 - val_acc: 0.7491
Epoch 51/100
390870/390870 [==============================] - 24s 61us/step - loss: 0.5146 - acc: 0.7513 - val_loss: 0.5201 - val_acc: 0.7514
Epoch 52/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5137 - acc: 0.7524 - val_loss: 0.5248 - val_acc: 0.7475
Epoch 53/100
390870/390870 [==============================] - 23s 60us/step - loss: 0.5141 - acc: 0.7523 - val_loss: 0.5217 - val_acc: 0.7475
Epoch 54/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5132 - acc: 0.7525 - val_loss: 0.5188 - val_acc: 0.7510
Epoch 55/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5134 - acc: 0.7523 - val_loss: 0.5183 - val_acc: 0.7523
Epoch 56/100
390870/390870 [==============================] - 23s 60us/step - loss: 0.5128 - acc: 0.7526 - val_loss: 0.5206 - val_acc: 0.7486
Epoch 57/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5130 - acc: 0.7525 - val_loss: 0.5402 - val_acc: 0.7370
Epoch 58/100
390870/390870 [==============================] - 24s 60us/step - loss: 0.5136 - acc: 0.7521 - val_loss: 0.5204 - val_acc: 0.7499
Epoch 59/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5135 - acc: 0.7517 - val_loss: 0.5186 - val_acc: 0.7510
Epoch 60/100
390870/390870 [==============================] - 24s 62us/step - loss: 0.5132 - acc: 0.7522 - val_loss: 0.5194 - val_acc: 0.7507
Epoch 61/100
390870/390870 [==============================] - 24s 61us/step - loss: 0.5131 - acc: 0.7523 - val_loss: 0.5266 - val_acc: 0.7503
Epoch 62/100
390870/390870 [==============================] - 24s 61us/step - loss: 0.5115 - acc: 0.7538 - val_loss: 0.5164 - val_acc: 0.7518
Epoch 63/100
390870/390870 [==============================] - 24s 61us/step - loss: 0.5140 - acc: 0.7522 - val_loss: 0.5255 - val_acc: 0.7511
Epoch 64/100
390870/390870 [==============================] - 24s 62us/step - loss: 0.5138 - acc: 0.7515 - val_loss: 0.5199 - val_acc: 0.7496
Epoch 65/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5141 - acc: 0.7517 - val_loss: 0.5218 - val_acc: 0.7502
Epoch 66/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5135 - acc: 0.7525 - val_loss: 0.5220 - val_acc: 0.7477
Epoch 67/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5152 - acc: 0.7519 - val_loss: 0.5243 - val_acc: 0.7497
Epoch 68/100
390870/390870 [==============================] - 25s 63us/step - loss: 0.5137 - acc: 0.7518 - val_loss: 0.5184 - val_acc: 0.7512
Epoch 69/100
390870/390870 [==============================] - 25s 63us/step - loss: 0.5174 - acc: 0.7515 - val_loss: 0.5292 - val_acc: 0.7474
Epoch 70/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5138 - acc: 0.7516 - val_loss: 0.5265 - val_acc: 0.7449
Epoch 71/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5127 - acc: 0.7534 - val_loss: 0.5263 - val_acc: 0.7479
Epoch 72/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5125 - acc: 0.7533 - val_loss: 0.5214 - val_acc: 0.7481
Epoch 73/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5119 - acc: 0.7531 - val_loss: 0.5209 - val_acc: 0.7503
Epoch 74/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5123 - acc: 0.7526 - val_loss: 0.5200 - val_acc: 0.7488
Epoch 75/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5109 - acc: 0.7536 - val_loss: 0.5184 - val_acc: 0.7513
Epoch 76/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5133 - acc: 0.7517 - val_loss: 0.5296 - val_acc: 0.7484
Epoch 77/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5128 - acc: 0.7520 - val_loss: 0.5254 - val_acc: 0.7467
Epoch 78/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5124 - acc: 0.7530 - val_loss: 0.5206 - val_acc: 0.7495
Epoch 79/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5120 - acc: 0.7534 - val_loss: 0.5209 - val_acc: 0.7489
Epoch 80/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5106 - acc: 0.7540 - val_loss: 0.5274 - val_acc: 0.7467
Epoch 81/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5137 - acc: 0.7519 - val_loss: 0.5233 - val_acc: 0.7467
Epoch 82/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5141 - acc: 0.7507 - val_loss: 0.5208 - val_acc: 0.7471
Epoch 83/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5109 - acc: 0.7538 - val_loss: 0.5199 - val_acc: 0.7525
Epoch 84/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5109 - acc: 0.7531 - val_loss: 0.5244 - val_acc: 0.7486
Epoch 85/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5119 - acc: 0.7528 - val_loss: 0.5202 - val_acc: 0.7500
Epoch 86/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5106 - acc: 0.7541 - val_loss: 0.5189 - val_acc: 0.7506
Epoch 87/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5110 - acc: 0.7535 - val_loss: 0.5202 - val_acc: 0.7517
Epoch 88/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5114 - acc: 0.7532 - val_loss: 0.5246 - val_acc: 0.7474
Epoch 89/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5105 - acc: 0.7543 - val_loss: 0.5199 - val_acc: 0.7481
Epoch 90/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5123 - acc: 0.7527 - val_loss: 0.5172 - val_acc: 0.7505
Epoch 91/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5119 - acc: 0.7523 - val_loss: 0.5270 - val_acc: 0.7457
Epoch 92/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5116 - acc: 0.7524 - val_loss: 0.5169 - val_acc: 0.7524
Epoch 93/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5094 - acc: 0.7547 - val_loss: 0.5164 - val_acc: 0.7524
Epoch 94/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5084 - acc: 0.7551 - val_loss: 0.5171 - val_acc: 0.7507
Epoch 95/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5084 - acc: 0.7552 - val_loss: 0.5224 - val_acc: 0.7516
Epoch 96/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5106 - acc: 0.7539 - val_loss: 0.5236 - val_acc: 0.7490
Epoch 97/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5095 - acc: 0.7540 - val_loss: 0.5454 - val_acc: 0.7475
Epoch 98/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5096 - acc: 0.7538 - val_loss: 0.5237 - val_acc: 0.7514
Epoch 99/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5099 - acc: 0.7541 - val_loss: 0.5264 - val_acc: 0.7490
Epoch 100/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5098 - acc: 0.7541 - val_loss: 0.5324 - val_acc: 0.7484

<keras.callbacks.History at 0x234bf5d0908>

# Save trained model
classifier.save("../models/trained_deep_neural_network.h5")

# evaluation of the model training
evaluation = classifier.evaluate(train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])
print(evaluation)

558386/558386 [==============================] - 16s 29us/step
[0.516120845819726, 0.7530937380230908]

The neural network gives us the probability that the class is 1, we binarize with threshold to 0.5 thus having the labels with probability up to 0.5 at 0, the others at 1. We could use two nerones in output thus having a probability vector expressing the probability that it is 1 and the probability that it is 0, but only an equivalent approach would result.

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_3.pickle')
test_set.head()

prediction = classifier.predict(test_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
binary_prediction = _binarize(prediction, threshold = 0.5)

print("Report for Deep Learning approach:")
print(_classification_report(test_set['likes'], binary_prediction))

print("Accuracy for Deep Learning approach:", _accuracy_score(test_set['likes'], binary_prediction) * 100)

Report for Deep Learning approach:
              precision    recall  f1-score   support

           0       0.69      0.40      0.51     50930
           1       0.75      0.91      0.83    103063

    accuracy                           0.74    153993
   macro avg       0.72      0.66      0.67    153993
weighted avg       0.73      0.74      0.72    153993

Accuracy for Deep Learning approach: 74.20402226075211

matrix = _confusion_matrix(test_set['likes'], binary_prediction)
print(matrix)

[[20454 30476]
 [ 9248 93815]]

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], binary_prediction)

_plt.plot(fpr, tpr)
_plt.xlim([0.0, 1.0])
_plt.ylim([0.0, 1.0])

_plt.title("Neural Network ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

NN 2¶

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')

# Number of neurons for hiddens layer
number_samples, number_features = train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']).shape

number_input_neurons = number_features
number_output_neurons = 1
number_train_semples = number_samples
alpha = 7

number_hidden_neurons = round(number_train_semples / (alpha * (number_input_neurons + number_output_neurons)))
print(number_hidden_neurons)

102

classifier = _Sequential()

# First Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal', input_dim = number_features))

# Second Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Third Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fourth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fifth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Output Layer
classifier.add(_Dense(1, activation = 'sigmoid', kernel_initializer = 'random_normal'))

# Compiling the neural network
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

# Fitting the data to the training dataset
classifier.fit(train_set.drop(
    columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']),
    train_set['likes'], validation_split = 0.3, batch_size = 100, epochs = 100)

Train on 390870 samples, validate on 167516 samples
Epoch 1/100
390870/390870 [==============================] - 36s 93us/step - loss: 0.5886 - acc: 0.6996 - val_loss: 0.5660 - val_acc: 0.7186
Epoch 2/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5544 - acc: 0.7243 - val_loss: 0.5464 - val_acc: 0.7330
Epoch 3/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5462 - acc: 0.7305 - val_loss: 0.5435 - val_acc: 0.7344
Epoch 4/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5417 - acc: 0.7337 - val_loss: 0.5363 - val_acc: 0.7391
Epoch 5/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5386 - acc: 0.7359 - val_loss: 0.5388 - val_acc: 0.7387
Epoch 6/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5362 - acc: 0.7374 - val_loss: 0.5397 - val_acc: 0.7345
Epoch 7/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5345 - acc: 0.7390 - val_loss: 0.5341 - val_acc: 0.7415
Epoch 8/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5333 - acc: 0.7401 - val_loss: 0.5340 - val_acc: 0.7381
Epoch 9/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5320 - acc: 0.7408 - val_loss: 0.5331 - val_acc: 0.7414
Epoch 10/100
390870/390870 [==============================] - 29s 75us/step - loss: 0.5308 - acc: 0.7418 - val_loss: 0.5435 - val_acc: 0.7391
Epoch 11/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5297 - acc: 0.7423 - val_loss: 0.5311 - val_acc: 0.7448
Epoch 12/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5290 - acc: 0.7427 - val_loss: 0.5284 - val_acc: 0.7451
Epoch 13/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5276 - acc: 0.7437 - val_loss: 0.5314 - val_acc: 0.7428
Epoch 14/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5272 - acc: 0.7441 - val_loss: 0.5279 - val_acc: 0.7445
Epoch 15/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5261 - acc: 0.7453 - val_loss: 0.5284 - val_acc: 0.7460
Epoch 16/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5258 - acc: 0.7452 - val_loss: 0.5294 - val_acc: 0.7438
Epoch 17/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5249 - acc: 0.7458 - val_loss: 0.5251 - val_acc: 0.7474
Epoch 18/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5241 - acc: 0.7461 - val_loss: 0.5286 - val_acc: 0.7455
Epoch 19/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5240 - acc: 0.7461 - val_loss: 0.5365 - val_acc: 0.7435
Epoch 20/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5240 - acc: 0.7461 - val_loss: 0.5237 - val_acc: 0.7488
Epoch 21/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5230 - acc: 0.7472 - val_loss: 0.5251 - val_acc: 0.7476
Epoch 22/100
390870/390870 [==============================] - 30s 77us/step - loss: 0.5225 - acc: 0.7470 - val_loss: 0.5434 - val_acc: 0.7404
Epoch 23/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5215 - acc: 0.7472 - val_loss: 0.5368 - val_acc: 0.7448
Epoch 24/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5214 - acc: 0.7477 - val_loss: 0.5253 - val_acc: 0.7475
Epoch 25/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5204 - acc: 0.7487 - val_loss: 0.5236 - val_acc: 0.7473
Epoch 26/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5209 - acc: 0.7479 - val_loss: 0.5234 - val_acc: 0.7487
Epoch 27/100
390870/390870 [==============================] - 30s 76us/step - loss: 0.5207 - acc: 0.7476 - val_loss: 0.5228 - val_acc: 0.7478
Epoch 28/100
390870/390870 [==============================] - 29s 75us/step - loss: 0.5205 - acc: 0.7486 - val_loss: 0.5221 - val_acc: 0.7495
Epoch 29/100
390870/390870 [==============================] - 29s 73us/step - loss: 0.5198 - acc: 0.7490 - val_loss: 0.5355 - val_acc: 0.7393
Epoch 30/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5194 - acc: 0.7486 - val_loss: 0.5244 - val_acc: 0.7478
Epoch 31/100
390870/390870 [==============================] - 29s 73us/step - loss: 0.5185 - acc: 0.7496 - val_loss: 0.5238 - val_acc: 0.7475
Epoch 32/100
390870/390870 [==============================] - 29s 73us/step - loss: 0.5182 - acc: 0.7494 - val_loss: 0.5210 - val_acc: 0.7490
Epoch 33/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5181 - acc: 0.7498 - val_loss: 0.5216 - val_acc: 0.7479
Epoch 34/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5170 - acc: 0.7502 - val_loss: 0.5240 - val_acc: 0.7456
Epoch 35/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5168 - acc: 0.7507 - val_loss: 0.5366 - val_acc: 0.7351
Epoch 36/100
390870/390870 [==============================] - 29s 74us/step - loss: 0.5170 - acc: 0.7507 - val_loss: 0.5230 - val_acc: 0.7468
Epoch 37/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5165 - acc: 0.7510 - val_loss: 0.5223 - val_acc: 0.7490
Epoch 38/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5163 - acc: 0.7510 - val_loss: 0.5275 - val_acc: 0.7460
Epoch 39/100
390870/390870 [==============================] - 30s 76us/step - loss: 0.5156 - acc: 0.7512 - val_loss: 0.5281 - val_acc: 0.7455
Epoch 40/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5157 - acc: 0.7510 - val_loss: 0.5248 - val_acc: 0.7486
Epoch 41/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5159 - acc: 0.7511 - val_loss: 0.5222 - val_acc: 0.7483
Epoch 42/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5156 - acc: 0.7510 - val_loss: 0.5300 - val_acc: 0.7438
Epoch 43/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5169 - acc: 0.7499 - val_loss: 0.5327 - val_acc: 0.7407
Epoch 44/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5151 - acc: 0.7513 - val_loss: 0.5237 - val_acc: 0.7489
Epoch 45/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5148 - acc: 0.7512 - val_loss: 0.5250 - val_acc: 0.7479
Epoch 46/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5146 - acc: 0.7514 - val_loss: 0.5200 - val_acc: 0.7504
Epoch 47/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5150 - acc: 0.7520 - val_loss: 0.5191 - val_acc: 0.7482
Epoch 48/100
390870/390870 [==============================] - 29s 74us/step - loss: 0.5149 - acc: 0.7515 - val_loss: 0.5229 - val_acc: 0.7511
Epoch 49/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5143 - acc: 0.7518 - val_loss: 0.5303 - val_acc: 0.7425
Epoch 50/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5139 - acc: 0.7525 - val_loss: 0.5248 - val_acc: 0.7477
Epoch 51/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5128 - acc: 0.7530 - val_loss: 0.5198 - val_acc: 0.7497
Epoch 52/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5138 - acc: 0.7517 - val_loss: 0.5212 - val_acc: 0.7498
Epoch 53/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5125 - acc: 0.7527 - val_loss: 0.5240 - val_acc: 0.7475
Epoch 54/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5129 - acc: 0.7526 - val_loss: 0.5239 - val_acc: 0.7477
Epoch 55/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5123 - acc: 0.7532 - val_loss: 0.5188 - val_acc: 0.7519
Epoch 56/100
390870/390870 [==============================] - 29s 74us/step - loss: 0.5122 - acc: 0.7531 - val_loss: 0.5258 - val_acc: 0.7484
Epoch 57/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5117 - acc: 0.7531 - val_loss: 0.5245 - val_acc: 0.7498
Epoch 58/100
390870/390870 [==============================] - 29s 73us/step - loss: 0.5121 - acc: 0.7530 - val_loss: 0.5228 - val_acc: 0.7505
Epoch 59/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5122 - acc: 0.7530 - val_loss: 0.5191 - val_acc: 0.7513
Epoch 60/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5113 - acc: 0.7537 - val_loss: 0.5171 - val_acc: 0.7529
Epoch 61/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5114 - acc: 0.7529 - val_loss: 0.5181 - val_acc: 0.7525
Epoch 62/100
390870/390870 [==============================] - 29s 74us/step - loss: 0.5112 - acc: 0.7539 - val_loss: 0.5203 - val_acc: 0.7495
Epoch 63/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5108 - acc: 0.7538 - val_loss: 0.5390 - val_acc: 0.7358
Epoch 64/100
390870/390870 [==============================] - 29s 74us/step - loss: 0.5104 - acc: 0.7538 - val_loss: 0.5204 - val_acc: 0.7518
Epoch 65/100
390870/390870 [==============================] - 29s 73us/step - loss: 0.5107 - acc: 0.7542 - val_loss: 0.5220 - val_acc: 0.7520
Epoch 66/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5102 - acc: 0.7547 - val_loss: 0.5327 - val_acc: 0.7485
Epoch 67/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5102 - acc: 0.7541 - val_loss: 0.5219 - val_acc: 0.7509
Epoch 68/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5099 - acc: 0.7545 - val_loss: 0.5195 - val_acc: 0.7510
Epoch 69/100
390870/390870 [==============================] - 29s 74us/step - loss: 0.5120 - acc: 0.7540 - val_loss: 0.5261 - val_acc: 0.7479
Epoch 70/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5105 - acc: 0.7543 - val_loss: 0.5225 - val_acc: 0.7490
Epoch 71/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5094 - acc: 0.7546 - val_loss: 0.5209 - val_acc: 0.7513
Epoch 72/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5095 - acc: 0.7550 - val_loss: 0.5199 - val_acc: 0.7501
Epoch 73/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5091 - acc: 0.7546 - val_loss: 0.5214 - val_acc: 0.7514
Epoch 74/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5100 - acc: 0.7540 - val_loss: 0.5497 - val_acc: 0.7407
Epoch 75/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5104 - acc: 0.7537 - val_loss: 0.5188 - val_acc: 0.7525
Epoch 76/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5090 - acc: 0.7547 - val_loss: 0.5196 - val_acc: 0.7532
Epoch 77/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5088 - acc: 0.7549 - val_loss: 0.5288 - val_acc: 0.7488
Epoch 78/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5086 - acc: 0.7548 - val_loss: 0.5233 - val_acc: 0.7520
Epoch 79/100
390870/390870 [==============================] - 29s 74us/step - loss: 0.5081 - acc: 0.7555 - val_loss: 0.5286 - val_acc: 0.7513
Epoch 80/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5089 - acc: 0.7548 - val_loss: 0.5277 - val_acc: 0.7494
Epoch 81/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5077 - acc: 0.7555 - val_loss: 0.5408 - val_acc: 0.7459
Epoch 82/100
390870/390870 [==============================] - 29s 75us/step - loss: 0.5085 - acc: 0.7553 - val_loss: 0.5278 - val_acc: 0.7526
Epoch 83/100
390870/390870 [==============================] - 29s 74us/step - loss: 0.5077 - acc: 0.7557 - val_loss: 0.5353 - val_acc: 0.7459
Epoch 84/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5072 - acc: 0.7560 - val_loss: 0.5307 - val_acc: 0.7504
Epoch 85/100
390870/390870 [==============================] - 29s 74us/step - loss: 0.5080 - acc: 0.7552 - val_loss: 0.5322 - val_acc: 0.7509
Epoch 86/100
390870/390870 [==============================] - 29s 73us/step - loss: 0.5071 - acc: 0.7557 - val_loss: 0.5326 - val_acc: 0.7517
Epoch 87/100
390870/390870 [==============================] - 29s 73us/step - loss: 0.5076 - acc: 0.7553 - val_loss: 0.5277 - val_acc: 0.7530
Epoch 88/100
390870/390870 [==============================] - 29s 74us/step - loss: 0.5075 - acc: 0.7559 - val_loss: 0.5375 - val_acc: 0.7482
Epoch 89/100
390870/390870 [==============================] - 30s 76us/step - loss: 0.5072 - acc: 0.7559 - val_loss: 0.5296 - val_acc: 0.7523
Epoch 90/100
390870/390870 [==============================] - 29s 75us/step - loss: 0.5063 - acc: 0.7566 - val_loss: 0.5287 - val_acc: 0.7526
Epoch 91/100
390870/390870 [==============================] - 30s 76us/step - loss: 0.5075 - acc: 0.7563 - val_loss: 0.5611 - val_acc: 0.7467
Epoch 92/100
390870/390870 [==============================] - 29s 74us/step - loss: 0.5070 - acc: 0.7558 - val_loss: 0.5295 - val_acc: 0.7526
Epoch 93/100
390870/390870 [==============================] - 29s 75us/step - loss: 0.5061 - acc: 0.7562 - val_loss: 0.5373 - val_acc: 0.7495
Epoch 94/100
390870/390870 [==============================] - 29s 73us/step - loss: 0.5068 - acc: 0.7561 - val_loss: 0.5299 - val_acc: 0.7536
Epoch 95/100
390870/390870 [==============================] - 29s 73us/step - loss: 0.5073 - acc: 0.7557 - val_loss: 0.5322 - val_acc: 0.7515
Epoch 96/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5057 - acc: 0.7571 - val_loss: 0.5339 - val_acc: 0.7524
Epoch 97/100
390870/390870 [==============================] - 29s 73us/step - loss: 0.5060 - acc: 0.7569 - val_loss: 0.5323 - val_acc: 0.7510
Epoch 98/100
390870/390870 [==============================] - 29s 73us/step - loss: 0.5065 - acc: 0.7561 - val_loss: 0.5350 - val_acc: 0.7505
Epoch 99/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5079 - acc: 0.7552 - val_loss: 0.5403 - val_acc: 0.7478
Epoch 100/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5072 - acc: 0.7562 - val_loss: 0.5282 - val_acc: 0.7527

<keras.callbacks.History at 0x20c82b579e8>

# Save trained model
classifier.save("../models/trained_deep_neural_network_2.h5")

# evaluation of the model training
evaluation = classifier.evaluate(train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])
print(evaluation)

558386/558386 [==============================] - 19s 35us/step
[0.5097503033514922, 0.7573470681567259]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_3.pickle')

prediction = classifier.predict(test_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
binary_prediction = _binarize(prediction, threshold = 0.5)

print("Report for Deep Learning approach:")
print(_classification_report(test_set['likes'], binary_prediction))

print("Accuracy for Deep Learning approach:", _accuracy_score(test_set['likes'], binary_prediction) * 100)

Report for Deep Learning approach:
              precision    recall  f1-score   support

           0       0.69      0.43      0.53     50930
           1       0.76      0.90      0.83    103063

    accuracy                           0.75    153993
   macro avg       0.72      0.67      0.68    153993
weighted avg       0.74      0.75      0.73    153993

Accuracy for Deep Learning approach: 74.67871916255933

matrix = _confusion_matrix(test_set['likes'], binary_prediction)
print(matrix)

[[22043 28887]
 [10106 92957]]

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], binary_prediction)

_plt.plot(fpr, tpr)
_plt.xlim([0.0, 1.0])
_plt.ylim([0.0, 1.0])

_plt.title("Neural Network ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

NN 3¶

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')

# Number of neurons for hiddens layer
number_samples, number_features = train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']).shape

number_input_neurons = number_features
number_output_neurons = 1
number_train_semples = number_samples
alpha = 8

number_hidden_neurons = round(number_train_semples / (alpha * (number_input_neurons + number_output_neurons)))
print(number_hidden_neurons)

89

classifier = _Sequential()

# First Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal', input_dim = number_features))

# Second Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Third Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fourth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fifth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Output Layer
classifier.add(_Dense(1, activation = 'sigmoid', kernel_initializer = 'random_normal'))

# Compiling the neural network
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

# Fitting the data to the training dataset
classifier.fit(train_set.drop(
    columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']),
    train_set['likes'], validation_split = 0.3, batch_size = 100, epochs = 100)

Train on 390870 samples, validate on 167516 samples
Epoch 1/100
390870/390870 [==============================] - 86s 220us/step - loss: 0.5889 - acc: 0.6991 - val_loss: 0.5599 - val_acc: 0.7175
Epoch 2/100
390870/390870 [==============================] - 80s 204us/step - loss: 0.5559 - acc: 0.7238 - val_loss: 0.5437 - val_acc: 0.7385
Epoch 3/100
390870/390870 [==============================] - 42s 109us/step - loss: 0.5473 - acc: 0.7302 - val_loss: 0.5678 - val_acc: 0.7257
Epoch 4/100
390870/390870 [==============================] - 26s 68us/step - loss: 0.5418 - acc: 0.7337 - val_loss: 0.5395 - val_acc: 0.7423
Epoch 5/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5385 - acc: 0.7362 - val_loss: 0.5365 - val_acc: 0.7395
Epoch 6/100
390870/390870 [==============================] - 26s 68us/step - loss: 0.5362 - acc: 0.7380 - val_loss: 0.5420 - val_acc: 0.7350
Epoch 7/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5351 - acc: 0.7386 - val_loss: 0.5403 - val_acc: 0.7368
Epoch 8/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5332 - acc: 0.7402 - val_loss: 0.5290 - val_acc: 0.7457
Epoch 9/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5317 - acc: 0.7414 - val_loss: 0.5361 - val_acc: 0.7418
Epoch 10/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5301 - acc: 0.7423 - val_loss: 0.5308 - val_acc: 0.7427
Epoch 11/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5298 - acc: 0.7424 - val_loss: 0.5374 - val_acc: 0.7386
Epoch 12/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5291 - acc: 0.7425 - val_loss: 0.5390 - val_acc: 0.7376
Epoch 13/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5269 - acc: 0.7444 - val_loss: 0.5296 - val_acc: 0.7429
Epoch 14/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5266 - acc: 0.7443 - val_loss: 0.5293 - val_acc: 0.7448
Epoch 15/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5254 - acc: 0.7455 - val_loss: 0.5251 - val_acc: 0.7473
Epoch 16/100
390870/390870 [==============================] - 26s 68us/step - loss: 0.5255 - acc: 0.7458 - val_loss: 0.5277 - val_acc: 0.7464
Epoch 17/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5240 - acc: 0.7467 - val_loss: 0.5280 - val_acc: 0.7478
Epoch 18/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5236 - acc: 0.7464 - val_loss: 0.5277 - val_acc: 0.7440
Epoch 19/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5227 - acc: 0.7474 - val_loss: 0.5282 - val_acc: 0.7450
Epoch 20/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5222 - acc: 0.7473 - val_loss: 0.5260 - val_acc: 0.7470
Epoch 21/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5218 - acc: 0.7481 - val_loss: 0.5247 - val_acc: 0.7461
Epoch 22/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5210 - acc: 0.7483 - val_loss: 0.5341 - val_acc: 0.7410
Epoch 23/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5211 - acc: 0.7482 - val_loss: 0.5279 - val_acc: 0.7444
Epoch 24/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5204 - acc: 0.7487 - val_loss: 0.5208 - val_acc: 0.7501
Epoch 25/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5200 - acc: 0.7488 - val_loss: 0.5300 - val_acc: 0.7450
Epoch 26/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5196 - acc: 0.7489 - val_loss: 0.5303 - val_acc: 0.7449
Epoch 27/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5192 - acc: 0.7493 - val_loss: 0.5473 - val_acc: 0.7363
Epoch 28/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5191 - acc: 0.7493 - val_loss: 0.5274 - val_acc: 0.7452
Epoch 29/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5182 - acc: 0.7502 - val_loss: 0.5204 - val_acc: 0.7489
Epoch 30/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5179 - acc: 0.7502 - val_loss: 0.5221 - val_acc: 0.7478
Epoch 31/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5179 - acc: 0.7500 - val_loss: 0.5256 - val_acc: 0.7455
Epoch 32/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5172 - acc: 0.7505 - val_loss: 0.5320 - val_acc: 0.7485
Epoch 33/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5165 - acc: 0.7511 - val_loss: 0.5248 - val_acc: 0.7486
Epoch 34/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5169 - acc: 0.7506 - val_loss: 0.5360 - val_acc: 0.7463
Epoch 35/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5160 - acc: 0.7514 - val_loss: 0.5229 - val_acc: 0.7469
Epoch 36/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5165 - acc: 0.7505 - val_loss: 0.5240 - val_acc: 0.7481
Epoch 37/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5160 - acc: 0.7511 - val_loss: 0.5360 - val_acc: 0.7463
Epoch 38/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5150 - acc: 0.7519 - val_loss: 0.5299 - val_acc: 0.7511
Epoch 39/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5147 - acc: 0.7517 - val_loss: 0.5296 - val_acc: 0.7458
Epoch 40/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5153 - acc: 0.7510 - val_loss: 0.5226 - val_acc: 0.7505
Epoch 41/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5142 - acc: 0.7523 - val_loss: 0.5264 - val_acc: 0.7489
Epoch 42/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5143 - acc: 0.7515 - val_loss: 0.5521 - val_acc: 0.7451
Epoch 43/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5144 - acc: 0.7522 - val_loss: 0.5299 - val_acc: 0.7487
Epoch 44/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5143 - acc: 0.7518 - val_loss: 0.5351 - val_acc: 0.7491
Epoch 45/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5137 - acc: 0.7525 - val_loss: 0.5300 - val_acc: 0.7493
Epoch 46/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5130 - acc: 0.7529 - val_loss: 0.5297 - val_acc: 0.7479
Epoch 47/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5134 - acc: 0.7528 - val_loss: 0.5308 - val_acc: 0.7460
Epoch 48/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5125 - acc: 0.7529 - val_loss: 0.5525 - val_acc: 0.7475
Epoch 49/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5125 - acc: 0.7532 - val_loss: 0.5977 - val_acc: 0.7437
Epoch 50/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5134 - acc: 0.7524 - val_loss: 0.5662 - val_acc: 0.7491
Epoch 51/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5125 - acc: 0.7535 - val_loss: 0.5283 - val_acc: 0.7511
Epoch 52/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5118 - acc: 0.7534 - val_loss: 0.5900 - val_acc: 0.7467
Epoch 53/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5117 - acc: 0.7534 - val_loss: 0.5814 - val_acc: 0.7486
Epoch 54/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5113 - acc: 0.7539 - val_loss: 0.5490 - val_acc: 0.7513
Epoch 55/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5109 - acc: 0.7541 - val_loss: 0.6131 - val_acc: 0.7465
Epoch 56/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5114 - acc: 0.7540 - val_loss: 0.6025 - val_acc: 0.7458
Epoch 57/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5107 - acc: 0.7537 - val_loss: 0.5912 - val_acc: 0.7473
Epoch 58/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5101 - acc: 0.7542 - val_loss: 0.6068 - val_acc: 0.7470
Epoch 59/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5125 - acc: 0.7535 - val_loss: 0.5276 - val_acc: 0.7503
Epoch 60/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5103 - acc: 0.7543 - val_loss: 0.5908 - val_acc: 0.7498
Epoch 61/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5100 - acc: 0.7544 - val_loss: 0.5929 - val_acc: 0.7458
Epoch 62/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5102 - acc: 0.7544 - val_loss: 0.5844 - val_acc: 0.7492
Epoch 63/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5099 - acc: 0.7542 - val_loss: 0.5892 - val_acc: 0.7468
Epoch 64/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5092 - acc: 0.7546 - val_loss: 0.5878 - val_acc: 0.7459
Epoch 65/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5099 - acc: 0.7552 - val_loss: 0.5823 - val_acc: 0.7465
Epoch 66/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5098 - acc: 0.7549 - val_loss: 0.5856 - val_acc: 0.7455
Epoch 67/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5095 - acc: 0.7547 - val_loss: 0.5800 - val_acc: 0.7489
Epoch 68/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5095 - acc: 0.7548 - val_loss: 0.6077 - val_acc: 0.7485
Epoch 69/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5085 - acc: 0.7552 - val_loss: 0.5824 - val_acc: 0.7474
Epoch 70/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5090 - acc: 0.7548 - val_loss: 0.5980 - val_acc: 0.7468
Epoch 71/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5091 - acc: 0.7549 - val_loss: 0.5952 - val_acc: 0.7473
Epoch 72/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5081 - acc: 0.7557 - val_loss: 0.6037 - val_acc: 0.7467
Epoch 73/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5097 - acc: 0.7551 - val_loss: 0.6016 - val_acc: 0.7459
Epoch 74/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5102 - acc: 0.7550 - val_loss: 0.5877 - val_acc: 0.7470
Epoch 75/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5084 - acc: 0.7553 - val_loss: 0.6011 - val_acc: 0.7427
Epoch 76/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5086 - acc: 0.7553 - val_loss: 0.5909 - val_acc: 0.7465
Epoch 77/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5080 - acc: 0.7553 - val_loss: 0.5588 - val_acc: 0.7485
Epoch 78/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5075 - acc: 0.7563 - val_loss: 0.5866 - val_acc: 0.7496
Epoch 79/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5079 - acc: 0.7554 - val_loss: 0.6126 - val_acc: 0.7452
Epoch 80/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5079 - acc: 0.7562 - val_loss: 0.5867 - val_acc: 0.7461
Epoch 81/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5076 - acc: 0.7558 - val_loss: 0.5865 - val_acc: 0.7499
Epoch 82/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5071 - acc: 0.7559 - val_loss: 0.5533 - val_acc: 0.7494
Epoch 83/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5078 - acc: 0.7559 - val_loss: 0.6015 - val_acc: 0.7485
Epoch 84/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5075 - acc: 0.7561 - val_loss: 0.5866 - val_acc: 0.7458
Epoch 85/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5075 - acc: 0.7557 - val_loss: 0.5920 - val_acc: 0.7487
Epoch 86/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5074 - acc: 0.7558 - val_loss: 0.5795 - val_acc: 0.7513
Epoch 87/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5064 - acc: 0.7567 - val_loss: 0.5217 - val_acc: 0.7531
Epoch 88/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5067 - acc: 0.7561 - val_loss: 0.5314 - val_acc: 0.7516
Epoch 89/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5061 - acc: 0.7565 - val_loss: 0.5377 - val_acc: 0.7501
Epoch 90/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5064 - acc: 0.7569 - val_loss: 0.5567 - val_acc: 0.7504
Epoch 91/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5060 - acc: 0.7568 - val_loss: 0.5890 - val_acc: 0.7495
Epoch 92/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5062 - acc: 0.7565 - val_loss: 0.5978 - val_acc: 0.7496
Epoch 93/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5063 - acc: 0.7569 - val_loss: 0.5849 - val_acc: 0.7479
Epoch 94/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5066 - acc: 0.7567 - val_loss: 0.5656 - val_acc: 0.7522
Epoch 95/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5055 - acc: 0.7570 - val_loss: 0.5727 - val_acc: 0.7496
Epoch 96/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5065 - acc: 0.7563 - val_loss: 0.5764 - val_acc: 0.7490
Epoch 97/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5064 - acc: 0.7568 - val_loss: 0.5726 - val_acc: 0.7529
Epoch 98/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5057 - acc: 0.7573 - val_loss: 0.5820 - val_acc: 0.7494
Epoch 99/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5068 - acc: 0.7562 - val_loss: 0.5329 - val_acc: 0.7543
Epoch 100/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5062 - acc: 0.7571 - val_loss: 0.5933 - val_acc: 0.7459

<keras.callbacks.History at 0x209bf585a20>

# Save trained model
classifier.save("../models/trained_deep_neural_network_3.h5")

# evaluation of the model training
evaluation = classifier.evaluate(train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])
print(evaluation)

558386/558386 [==============================] - 18s 31us/step
[0.5327735302538145, 0.7514568775011509]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_3.pickle')

prediction = classifier.predict(test_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
binary_prediction = _binarize(prediction, threshold = 0.5)

print("Report for Deep Learning approach:")
print(_classification_report(test_set['likes'], binary_prediction))

print("Accuracy for Deep Learning approach:", _accuracy_score(test_set['likes'], binary_prediction) * 100)

Report for Deep Learning approach:
              precision    recall  f1-score   support

           0       0.75      0.31      0.44     50930
           1       0.74      0.95      0.83    103063

    accuracy                           0.74    153993
   macro avg       0.75      0.63      0.63    153993
weighted avg       0.74      0.74      0.70    153993

Accuracy for Deep Learning approach: 73.78582143344178

matrix = _confusion_matrix(test_set['likes'], binary_prediction)
print(matrix)

[[15639 35291]
 [ 5077 97986]]

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], binary_prediction)

_plt.plot(fpr, tpr)
_plt.xlim([0.0, 1.0])
_plt.ylim([0.0, 1.0])

_plt.title("Neural Network ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_del_all()

NN 4¶

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')

# Number of neurons for hiddens layer
number_samples, number_features = train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']).shape

number_input_neurons = number_features
number_output_neurons = 1
number_train_semples = number_samples
alpha = 8

number_hidden_neurons = round(number_train_semples / (alpha * (number_input_neurons + number_output_neurons)))
print(number_hidden_neurons)

89

classifier = _Sequential()

# First Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal', input_dim = number_features))

# Second Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Third Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fourth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fifth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Sixth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Output Layer
classifier.add(_Dense(1, activation = 'sigmoid', kernel_initializer = 'random_normal'))

# Compiling the neural network
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

# Fitting the data to the training dataset
classifier.fit(train_set.drop(
    columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']),
    train_set['likes'], validation_split = 0.3, batch_size = 100, epochs = 100)

Train on 390870 samples, validate on 167516 samples
Epoch 1/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5898 - acc: 0.6983 - val_loss: 0.5598 - val_acc: 0.7252
Epoch 2/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5545 - acc: 0.7250 - val_loss: 0.5465 - val_acc: 0.7327
Epoch 3/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5461 - acc: 0.7305 - val_loss: 0.5469 - val_acc: 0.7400
Epoch 4/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5418 - acc: 0.7335 - val_loss: 0.5371 - val_acc: 0.7366
Epoch 5/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5390 - acc: 0.7357 - val_loss: 0.5364 - val_acc: 0.7400
Epoch 6/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5367 - acc: 0.7374 - val_loss: 0.5425 - val_acc: 0.7357
Epoch 7/100
390870/390870 [==============================] - 26s 68us/step - loss: 0.5349 - acc: 0.7387 - val_loss: 0.5291 - val_acc: 0.7436
Epoch 8/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5325 - acc: 0.7404 - val_loss: 0.5406 - val_acc: 0.7351
Epoch 9/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5315 - acc: 0.7407 - val_loss: 0.5321 - val_acc: 0.7419
Epoch 10/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5297 - acc: 0.7426 - val_loss: 0.5471 - val_acc: 0.7369
Epoch 11/100
390870/390870 [==============================] - 26s 68us/step - loss: 0.5290 - acc: 0.7426 - val_loss: 0.5304 - val_acc: 0.7417
Epoch 12/100
390870/390870 [==============================] - 26s 68us/step - loss: 0.5283 - acc: 0.7427 - val_loss: 0.5318 - val_acc: 0.7437
Epoch 13/100
390870/390870 [==============================] - 26s 68us/step - loss: 0.5276 - acc: 0.7440 - val_loss: 0.5271 - val_acc: 0.7466
Epoch 14/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5264 - acc: 0.7448 - val_loss: 0.5311 - val_acc: 0.7446
Epoch 15/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5254 - acc: 0.7448 - val_loss: 0.5278 - val_acc: 0.7474
Epoch 16/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5251 - acc: 0.7452 - val_loss: 0.5250 - val_acc: 0.7486
Epoch 17/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5242 - acc: 0.7460 - val_loss: 0.5246 - val_acc: 0.7486
Epoch 18/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5229 - acc: 0.7475 - val_loss: 0.5234 - val_acc: 0.7495
Epoch 19/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5225 - acc: 0.7470 - val_loss: 0.5234 - val_acc: 0.7487
Epoch 20/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5224 - acc: 0.7473 - val_loss: 0.5548 - val_acc: 0.7346
Epoch 21/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5216 - acc: 0.7474 - val_loss: 0.5364 - val_acc: 0.7386
Epoch 22/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5214 - acc: 0.7478 - val_loss: 0.5233 - val_acc: 0.7467
Epoch 23/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5210 - acc: 0.7483 - val_loss: 0.5213 - val_acc: 0.7495
Epoch 24/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5201 - acc: 0.7489 - val_loss: 0.5240 - val_acc: 0.7486
Epoch 25/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5194 - acc: 0.7488 - val_loss: 0.5228 - val_acc: 0.7492
Epoch 26/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5182 - acc: 0.7494 - val_loss: 0.5199 - val_acc: 0.7498
Epoch 27/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5180 - acc: 0.7495 - val_loss: 0.5270 - val_acc: 0.7470
Epoch 28/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5185 - acc: 0.7491 - val_loss: 0.5269 - val_acc: 0.7466
Epoch 29/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5177 - acc: 0.7499 - val_loss: 0.5202 - val_acc: 0.7481
Epoch 30/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5168 - acc: 0.7499 - val_loss: 0.5275 - val_acc: 0.7441
Epoch 31/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5167 - acc: 0.7501 - val_loss: 0.5252 - val_acc: 0.7460
Epoch 32/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5163 - acc: 0.7510 - val_loss: 0.5395 - val_acc: 0.7435
Epoch 33/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5161 - acc: 0.7508 - val_loss: 0.5297 - val_acc: 0.7454
Epoch 34/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5165 - acc: 0.7509 - val_loss: 0.5225 - val_acc: 0.7504
Epoch 35/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5154 - acc: 0.7511 - val_loss: 0.5214 - val_acc: 0.7497
Epoch 36/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5149 - acc: 0.7516 - val_loss: 0.5183 - val_acc: 0.7519
Epoch 37/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5157 - acc: 0.7517 - val_loss: 0.5268 - val_acc: 0.7462
Epoch 38/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5141 - acc: 0.7521 - val_loss: 0.5173 - val_acc: 0.7523
Epoch 39/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5141 - acc: 0.7523 - val_loss: 0.5186 - val_acc: 0.7506
Epoch 40/100
390870/390870 [==============================] - 26s 68us/step - loss: 0.5143 - acc: 0.7517 - val_loss: 0.5218 - val_acc: 0.7516
Epoch 41/100
390870/390870 [==============================] - 26s 68us/step - loss: 0.5143 - acc: 0.7524 - val_loss: 0.5245 - val_acc: 0.7493
Epoch 42/100
390870/390870 [==============================] - 26s 68us/step - loss: 0.5132 - acc: 0.7526 - val_loss: 0.5260 - val_acc: 0.7487
Epoch 43/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5131 - acc: 0.7527 - val_loss: 0.5201 - val_acc: 0.7504
Epoch 44/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5132 - acc: 0.7531 - val_loss: 0.5354 - val_acc: 0.7408
Epoch 45/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5131 - acc: 0.7529 - val_loss: 0.5222 - val_acc: 0.7517
Epoch 46/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5124 - acc: 0.7536 - val_loss: 0.5339 - val_acc: 0.7439
Epoch 47/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5122 - acc: 0.7536 - val_loss: 0.5231 - val_acc: 0.7508
Epoch 48/100
390870/390870 [==============================] - 30s 77us/step - loss: 0.5116 - acc: 0.7534 - val_loss: 0.5285 - val_acc: 0.7508
Epoch 49/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5121 - acc: 0.7542 - val_loss: 0.5260 - val_acc: 0.7487
Epoch 50/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5114 - acc: 0.7538 - val_loss: 0.5295 - val_acc: 0.7500
Epoch 51/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5112 - acc: 0.7542 - val_loss: 0.5229 - val_acc: 0.7492
Epoch 52/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5108 - acc: 0.7535 - val_loss: 0.5436 - val_acc: 0.7470
Epoch 53/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5105 - acc: 0.7542 - val_loss: 0.5274 - val_acc: 0.7512
Epoch 54/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5107 - acc: 0.7547 - val_loss: 0.5283 - val_acc: 0.7512
Epoch 55/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5099 - acc: 0.7543 - val_loss: 0.5323 - val_acc: 0.7503
Epoch 56/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5102 - acc: 0.7545 - val_loss: 0.5280 - val_acc: 0.7497
Epoch 57/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5105 - acc: 0.7538 - val_loss: 0.5201 - val_acc: 0.7511
Epoch 58/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5096 - acc: 0.7543 - val_loss: 0.5202 - val_acc: 0.7525
Epoch 59/100
390870/390870 [==============================] - 30s 78us/step - loss: 0.5103 - acc: 0.7541 - val_loss: 0.5213 - val_acc: 0.7499
Epoch 60/100
390870/390870 [==============================] - 29s 73us/step - loss: 0.5098 - acc: 0.7546 - val_loss: 0.5211 - val_acc: 0.7524
Epoch 61/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5091 - acc: 0.7549 - val_loss: 0.5230 - val_acc: 0.7530
Epoch 62/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5095 - acc: 0.7550 - val_loss: 0.5298 - val_acc: 0.7495
Epoch 63/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5094 - acc: 0.7550 - val_loss: 0.5252 - val_acc: 0.7515
Epoch 64/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5096 - acc: 0.7550 - val_loss: 0.5241 - val_acc: 0.7513
Epoch 65/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5093 - acc: 0.7556 - val_loss: 0.5284 - val_acc: 0.7477
Epoch 66/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5084 - acc: 0.7558 - val_loss: 0.5233 - val_acc: 0.7527
Epoch 67/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5080 - acc: 0.7558 - val_loss: 0.5339 - val_acc: 0.7475
Epoch 68/100
390870/390870 [==============================] - 28s 73us/step - loss: 0.5081 - acc: 0.7553 - val_loss: 0.5362 - val_acc: 0.7507
Epoch 69/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5075 - acc: 0.7560 - val_loss: 0.5286 - val_acc: 0.7505
Epoch 70/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5081 - acc: 0.7555 - val_loss: 0.5277 - val_acc: 0.7513
Epoch 71/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5080 - acc: 0.7554 - val_loss: 0.5272 - val_acc: 0.7519
Epoch 72/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5080 - acc: 0.7558 - val_loss: 0.5238 - val_acc: 0.7527
Epoch 73/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5073 - acc: 0.7561 - val_loss: 0.5253 - val_acc: 0.7505
Epoch 74/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5075 - acc: 0.7557 - val_loss: 0.5201 - val_acc: 0.7533
Epoch 75/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5071 - acc: 0.7563 - val_loss: 0.5419 - val_acc: 0.7488
Epoch 76/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5077 - acc: 0.7563 - val_loss: 0.5243 - val_acc: 0.7527
Epoch 77/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5071 - acc: 0.7565 - val_loss: 0.5257 - val_acc: 0.7501
Epoch 78/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5071 - acc: 0.7559 - val_loss: 0.5348 - val_acc: 0.7487
Epoch 79/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5078 - acc: 0.7558 - val_loss: 0.5285 - val_acc: 0.7524
Epoch 80/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5066 - acc: 0.7567 - val_loss: 0.5261 - val_acc: 0.7530
Epoch 81/100
390870/390870 [==============================] - 28s 70us/step - loss: 0.5072 - acc: 0.7567 - val_loss: 0.5288 - val_acc: 0.7513
Epoch 82/100
390870/390870 [==============================] - 28s 70us/step - loss: 0.5060 - acc: 0.7572 - val_loss: 0.5311 - val_acc: 0.7516
Epoch 83/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5064 - acc: 0.7565 - val_loss: 0.5298 - val_acc: 0.7528
Epoch 84/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5066 - acc: 0.7563 - val_loss: 0.5271 - val_acc: 0.7517
Epoch 85/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5062 - acc: 0.7566 - val_loss: 0.5382 - val_acc: 0.7484
Epoch 86/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5065 - acc: 0.7566 - val_loss: 0.5254 - val_acc: 0.7529
Epoch 87/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5064 - acc: 0.7570 - val_loss: 0.5450 - val_acc: 0.7492
Epoch 88/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5056 - acc: 0.7574 - val_loss: 0.5294 - val_acc: 0.7522
Epoch 89/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5057 - acc: 0.7574 - val_loss: 0.5288 - val_acc: 0.7514
Epoch 90/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5058 - acc: 0.7570 - val_loss: 0.5240 - val_acc: 0.7525
Epoch 91/100
390870/390870 [==============================] - 29s 75us/step - loss: 0.5058 - acc: 0.7568 - val_loss: 0.5285 - val_acc: 0.7534
Epoch 92/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5055 - acc: 0.7573 - val_loss: 0.5356 - val_acc: 0.7483
Epoch 93/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5050 - acc: 0.7571 - val_loss: 0.5573 - val_acc: 0.7509
Epoch 94/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5052 - acc: 0.7575 - val_loss: 0.5331 - val_acc: 0.7510
Epoch 95/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5054 - acc: 0.7574 - val_loss: 0.5303 - val_acc: 0.7522
Epoch 96/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5050 - acc: 0.7578 - val_loss: 0.5299 - val_acc: 0.7520
Epoch 97/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5047 - acc: 0.7578 - val_loss: 0.5341 - val_acc: 0.7511
Epoch 98/100
390870/390870 [==============================] - 29s 73us/step - loss: 0.5051 - acc: 0.7577 - val_loss: 0.5604 - val_acc: 0.7486
Epoch 99/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5057 - acc: 0.7576 - val_loss: 0.5413 - val_acc: 0.7446
Epoch 100/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5045 - acc: 0.7576 - val_loss: 0.5278 - val_acc: 0.7536

<keras.callbacks.History at 0x209e032a048>

# Save trained model
classifier.save("../models/trained_deep_neural_network_4.h5")

# evaluation of the model training
evaluation = classifier.evaluate(train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])
print(evaluation)

558386/558386 [==============================] - 20s 35us/step
[0.5089105000514885, 0.758450247677916]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_3.pickle')

prediction = classifier.predict(test_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
binary_prediction = _binarize(prediction, threshold = 0.5)

print("Report for Deep Learning approach:")
print(_classification_report(test_set['likes'], binary_prediction))

print("Accuracy for Deep Learning approach:", _accuracy_score(test_set['likes'], binary_prediction) * 100)

Report for Deep Learning approach:
              precision    recall  f1-score   support

           0       0.70      0.41      0.52     50930
           1       0.76      0.91      0.83    103063

    accuracy                           0.75    153993
   macro avg       0.73      0.66      0.67    153993
weighted avg       0.74      0.75      0.72    153993

Accuracy for Deep Learning approach: 74.58455903839784

matrix = _confusion_matrix(test_set['likes'], binary_prediction)
print(matrix)

[[20848 30082]
 [ 9056 94007]]

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], binary_prediction)

_plt.plot(fpr, tpr)
_plt.xlim([0.0, 1.0])
_plt.ylim([0.0, 1.0])

_plt.title("Neural Network ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

7. Other experiments¶

7.1 SVM without dimensionality reduction¶

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_2.pickle')
train_set.shape

(558386, 2774)

best_model = _jl.load("../models/best_SVM.joblib")
best_model.set_params(verbose=10)
best_model.get_params()

{'C': 0.001,
 'class_weight': None,
 'dual': True,
 'fit_intercept': True,
 'intercept_scaling': 1,
 'loss': 'squared_hinge',
 'max_iter': 50000,
 'multi_class': 'ovr',
 'penalty': 'l2',
 'random_state': 0,
 'tol': 0.0001,
 'verbose': 10}

best_model.fit(train_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])

[LibLinear]

C:\Users\super\Anaconda3\envs\tf-gpu\lib\site-packages\sklearn\svm\base.py:929: ConvergenceWarning: Liblinear failed to converge, increase the number of iterations.
  "the number of iterations.", ConvergenceWarning)

LinearSVC(C=0.001, class_weight=None, dual=True, fit_intercept=True,
          intercept_scaling=1, loss='squared_hinge', max_iter=50000,
          multi_class='ovr', penalty='l2', random_state=0, tol=0.0001,
          verbose=10)

_jl.dump(best_model, "../models/best_SVM_all.joblib")

['../models/best_SVM_all.joblib']

print("coef:", best_model.coef_)
print("intercept:", best_model.intercept_)

coef: [[-0.14394614 -0.20443318  0.34380749 ...  0.00380986  0.00857759
   0.        ]]
intercept: [-0.31154943]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_2.pickle')
test_set.shape

(153993, 2774)

# test classifier
predic = best_model.predict(test_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
print("predictions:\n", predic)

predictions:
 [1 0 1 ... 0 0 0]

set(predic)

{0, 1}

# evaluate classifier

print("Report for Support Vector Machine:")
print(_classification_report(test_set['likes'], predic))

print("Accuracy for Support Vector Machine:", _accuracy_score(test_set['likes'], predic)*100)

Report for Support Vector Machine:
              precision    recall  f1-score   support

           0       0.72      0.34      0.46     50930
           1       0.74      0.93      0.83    103063

    accuracy                           0.74    153993
   macro avg       0.73      0.64      0.64    153993
weighted avg       0.73      0.74      0.71    153993

Accuracy for Support Vector Machine: 73.73971544161098

# Confusion matrix for SVC

print("Confusion Matrix for SVC: ")
_confusion_matrix(test_set['likes'], predic)

Confusion Matrix for SVC:

array([[17293, 33637],
       [ 6802, 96261]], dtype=int64)

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], predic)

_plt.plot(fpr,tpr)
_plt.xlim([0.0,1.0])
_plt.ylim([0.0,1.0])

_plt.title("SVM ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

7.2 Random forest without dimensionality reduction¶

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_2.pickle')
train_set.shape

(558386, 2774)

params = _jl.load("../models/best_Random_Forest_2.joblib").get_params()
params['n_jobs'] = -1
params['verbose'] = 10
best_model = _RandomForestClassifier(**params)
best_model.get_params()

{'bootstrap': False,
 'class_weight': None,
 'criterion': 'entropy',
 'max_depth': 50,
 'max_features': 'auto',
 'max_leaf_nodes': None,
 'min_impurity_decrease': 0.0,
 'min_impurity_split': None,
 'min_samples_leaf': 2,
 'min_samples_split': 10,
 'min_weight_fraction_leaf': 0.0,
 'n_estimators': 1000,
 'n_jobs': -1,
 'oob_score': False,
 'random_state': None,
 'verbose': 10,
 'warm_start': False}

best_model.fit(train_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])

[Parallel(n_jobs=-1)]: Using backend ThreadingBackend with 12 concurrent workers.

building tree 1 of 1000
building tree 2 of 1000
building tree 3 of 1000
building tree 4 of 1000
building tree 5 of 1000
building tree 6 of 1000
building tree 7 of 1000
building tree 8 of 1000
building tree 9 of 1000
building tree 10 of 1000
building tree 11 of 1000
building tree 12 of 1000

[Parallel(n_jobs=-1)]: Done   1 tasks      | elapsed:   44.0s

building tree 13 of 1000
building tree 14 of 1000
building tree 15 of 1000
building tree 16 of 1000
building tree 17 of 1000
building tree 18 of 1000
building tree 19 of 1000

[Parallel(n_jobs=-1)]: Done   8 tasks      | elapsed:   53.2s

building tree 20 of 1000
building tree 21 of 1000
building tree 22 of 1000
building tree 23 of 1000
building tree 24 of 1000
building tree 25 of 1000
building tree 26 of 1000
building tree 27 of 1000
building tree 28 of 1000

[Parallel(n_jobs=-1)]: Done  17 tasks      | elapsed:  1.9min

building tree 29 of 1000
building tree 30 of 1000
building tree 31 of 1000
building tree 32 of 1000
building tree 33 of 1000
building tree 34 of 1000
building tree 35 of 1000
building tree 36 of 1000
building tree 37 of 1000

[Parallel(n_jobs=-1)]: Done  26 tasks      | elapsed:  2.7min

building tree 38 of 1000
building tree 39 of 1000
building tree 40 of 1000
building tree 41 of 1000
building tree 42 of 1000
building tree 43 of 1000
building tree 44 of 1000
building tree 45 of 1000
building tree 46 of 1000
building tree 47 of 1000
building tree 48 of 1000

_jl.dump(best_model, "../models/best_Random_Forest_all.joblib")

['../models/best_Random_Forest_all.joblib']

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_2.pickle')
test_set.shape

(153993, 2774)

# test classifier
predic = best_model.predict(test_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
print("predictions:\n", predic)

[Parallel(n_jobs=12)]: Using backend ThreadingBackend with 12 concurrent workers.
[Parallel(n_jobs=12)]: Done   1 tasks      | elapsed:    0.4s
[Parallel(n_jobs=12)]: Done   8 tasks      | elapsed:    0.4s
[Parallel(n_jobs=12)]: Done  17 tasks      | elapsed:    0.8s
[Parallel(n_jobs=12)]: Done  26 tasks      | elapsed:    1.2s
[Parallel(n_jobs=12)]: Done  37 tasks      | elapsed:    1.5s
[Parallel(n_jobs=12)]: Done  48 tasks      | elapsed:    1.8s
[Parallel(n_jobs=12)]: Done  61 tasks      | elapsed:    2.5s
[Parallel(n_jobs=12)]: Done  74 tasks      | elapsed:    3.0s
[Parallel(n_jobs=12)]: Done  89 tasks      | elapsed:    3.4s
[Parallel(n_jobs=12)]: Done 104 tasks      | elapsed:    3.9s
[Parallel(n_jobs=12)]: Done 121 tasks      | elapsed:    4.5s
[Parallel(n_jobs=12)]: Done 138 tasks      | elapsed:    5.0s
[Parallel(n_jobs=12)]: Done 157 tasks      | elapsed:    5.7s
[Parallel(n_jobs=12)]: Done 176 tasks      | elapsed:    6.3s
[Parallel(n_jobs=12)]: Done 197 tasks      | elapsed:    7.0s
[Parallel(n_jobs=12)]: Done 218 tasks      | elapsed:    7.7s
[Parallel(n_jobs=12)]: Done 241 tasks      | elapsed:    8.5s
[Parallel(n_jobs=12)]: Done 264 tasks      | elapsed:    9.2s
[Parallel(n_jobs=12)]: Done 289 tasks      | elapsed:   10.1s
[Parallel(n_jobs=12)]: Done 314 tasks      | elapsed:   10.9s
[Parallel(n_jobs=12)]: Done 341 tasks      | elapsed:   11.7s
[Parallel(n_jobs=12)]: Done 368 tasks      | elapsed:   12.8s
[Parallel(n_jobs=12)]: Done 397 tasks      | elapsed:   14.3s
[Parallel(n_jobs=12)]: Done 426 tasks      | elapsed:   15.7s
[Parallel(n_jobs=12)]: Done 457 tasks      | elapsed:   17.1s
[Parallel(n_jobs=12)]: Done 488 tasks      | elapsed:   18.5s
[Parallel(n_jobs=12)]: Done 521 tasks      | elapsed:   19.8s
[Parallel(n_jobs=12)]: Done 554 tasks      | elapsed:   21.1s
[Parallel(n_jobs=12)]: Done 589 tasks      | elapsed:   22.4s
[Parallel(n_jobs=12)]: Done 624 tasks      | elapsed:   23.9s
[Parallel(n_jobs=12)]: Done 661 tasks      | elapsed:   25.4s
[Parallel(n_jobs=12)]: Done 698 tasks      | elapsed:   26.9s
[Parallel(n_jobs=12)]: Done 737 tasks      | elapsed:   28.4s
[Parallel(n_jobs=12)]: Done 776 tasks      | elapsed:   30.0s
[Parallel(n_jobs=12)]: Done 817 tasks      | elapsed:   31.5s
[Parallel(n_jobs=12)]: Done 858 tasks      | elapsed:   33.0s
[Parallel(n_jobs=12)]: Done 901 tasks      | elapsed:   34.5s
[Parallel(n_jobs=12)]: Done 944 tasks      | elapsed:   36.1s
[Parallel(n_jobs=12)]: Done 1000 out of 1000 | elapsed:   37.9s finished

predictions:
 [0 0 1 ... 0 1 0]

set(predic)

{0, 1}

# evaluate classifier

print("Report for Random Forest classifier:")
print(_classification_report(test_set['likes'], predic))

print("Accuracy for Random Forest Classifier:", _accuracy_score(test_set['likes'], predic)*100)

Report for Random Forest classifier:
              precision    recall  f1-score   support

           0       0.69      0.38      0.49     50930
           1       0.75      0.92      0.82    103063

    accuracy                           0.74    153993
   macro avg       0.72      0.65      0.66    153993
weighted avg       0.73      0.74      0.71    153993

Accuracy for Random Forest Classifier: 73.88907287993611

# Confusion matrix for Random Forest

print("Confusion Matrix for Random Forest: ")
_confusion_matrix(test_set['likes'], predic)

Confusion Matrix for Random Forest:

array([[19194, 31736],
       [ 8473, 94590]], dtype=int64)

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], predic)

_plt.plot(fpr,tpr)
_plt.xlim([0.0,1.0])
_plt.ylim([0.0,1.0])

_plt.title("Random Forest ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

7.3 Deep Learning without dimensionality reduction¶

NN1¶

alpha = 6

# hidden layers = 3

batch size = 100

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_2.pickle')
train_set.shape

(558386, 2774)

# Number of neurons for hiddens layer
number_samples, number_features = train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']).shape

number_input_neurons = number_features
number_output_neurons = 1
number_train_semples = number_samples
alpha = 6

number_hidden_neurons = round(number_train_semples / (alpha * (number_input_neurons + number_output_neurons)))
print(number_hidden_neurons)

34

classifier = _Sequential()

# First Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal', input_dim = number_features))

# Second Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Third Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Output Layer
classifier.add(_Dense(1, activation = 'sigmoid', kernel_initializer = 'random_normal'))

# Compiling the neural network
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

# Fitting the data to the training dataset
classifier.fit(train_set.drop(
    columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']),
    train_set['likes'], validation_split = 0.3, batch_size = 100, epochs = 100)

Train on 390870 samples, validate on 167516 samples
Epoch 1/100
390870/390870 [==============================] - 180s 461us/step - loss: 0.5932 - acc: 0.6968 - val_loss: 0.5693 - val_acc: 0.7170
Epoch 2/100
390870/390870 [==============================] - 143s 366us/step - loss: 0.5560 - acc: 0.7226 - val_loss: 0.5513 - val_acc: 0.7321
Epoch 3/100
390870/390870 [==============================] - 154s 395us/step - loss: 0.5464 - acc: 0.7302 - val_loss: 0.5414 - val_acc: 0.7336
Epoch 4/100
390870/390870 [==============================] - 131s 336us/step - loss: 0.5406 - acc: 0.7342 - val_loss: 0.5370 - val_acc: 0.7369
Epoch 5/100
390870/390870 [==============================] - 148s 379us/step - loss: 0.5383 - acc: 0.7357 - val_loss: 0.5326 - val_acc: 0.7431
Epoch 6/100
390870/390870 [==============================] - 163s 418us/step - loss: 0.5355 - acc: 0.7381 - val_loss: 0.5380 - val_acc: 0.7398
Epoch 7/100
390870/390870 [==============================] - 165s 422us/step - loss: 0.5340 - acc: 0.7391 - val_loss: 0.5382 - val_acc: 0.7412
Epoch 8/100
390870/390870 [==============================] - 164s 421us/step - loss: 0.5336 - acc: 0.7393 - val_loss: 0.5378 - val_acc: 0.7378
Epoch 9/100
390870/390870 [==============================] - 162s 414us/step - loss: 0.5312 - acc: 0.7413 - val_loss: 0.5396 - val_acc: 0.7402
Epoch 10/100
390870/390870 [==============================] - 165s 423us/step - loss: 0.5299 - acc: 0.7419 - val_loss: 0.5340 - val_acc: 0.7426
Epoch 11/100
390870/390870 [==============================] - 148s 380us/step - loss: 0.5294 - acc: 0.7427 - val_loss: 0.5297 - val_acc: 0.7432
Epoch 12/100
390870/390870 [==============================] - 160s 410us/step - loss: 0.5273 - acc: 0.7441 - val_loss: 0.5437 - val_acc: 0.7336
Epoch 13/100
390870/390870 [==============================] - 158s 405us/step - loss: 0.5268 - acc: 0.7442 - val_loss: 0.5254 - val_acc: 0.7457
Epoch 14/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5254 - acc: 0.7456 - val_loss: 0.5336 - val_acc: 0.7410
Epoch 15/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5254 - acc: 0.7451 - val_loss: 0.5256 - val_acc: 0.7471
Epoch 16/100
390870/390870 [==============================] - 68s 173us/step - loss: 0.5242 - acc: 0.7462 - val_loss: 0.5331 - val_acc: 0.7412
Epoch 17/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5242 - acc: 0.7461 - val_loss: 0.5247 - val_acc: 0.7477
Epoch 18/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5228 - acc: 0.7475 - val_loss: 0.5269 - val_acc: 0.7450
Epoch 19/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5239 - acc: 0.7469 - val_loss: 0.5239 - val_acc: 0.7480
Epoch 20/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5225 - acc: 0.7471 - val_loss: 0.5273 - val_acc: 0.7458
Epoch 21/100
390870/390870 [==============================] - 68s 173us/step - loss: 0.5220 - acc: 0.7477 - val_loss: 0.5330 - val_acc: 0.7401
Epoch 22/100
390870/390870 [==============================] - 68s 173us/step - loss: 0.5199 - acc: 0.7493 - val_loss: 0.5249 - val_acc: 0.7474
Epoch 23/100
390870/390870 [==============================] - 69s 176us/step - loss: 0.5196 - acc: 0.7490 - val_loss: 0.5430 - val_acc: 0.7456
Epoch 24/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5185 - acc: 0.7495 - val_loss: 0.5225 - val_acc: 0.7485
Epoch 25/100
390870/390870 [==============================] - 63s 161us/step - loss: 0.5180 - acc: 0.7501 - val_loss: 0.5236 - val_acc: 0.7483
Epoch 26/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5194 - acc: 0.7494 - val_loss: 0.5216 - val_acc: 0.7493
Epoch 27/100
390870/390870 [==============================] - 68s 175us/step - loss: 0.5184 - acc: 0.7498 - val_loss: 0.5272 - val_acc: 0.7468
Epoch 28/100
390870/390870 [==============================] - 71s 182us/step - loss: 0.5172 - acc: 0.7507 - val_loss: 0.5282 - val_acc: 0.7452
Epoch 29/100
390870/390870 [==============================] - 71s 181us/step - loss: 0.5174 - acc: 0.7503 - val_loss: 0.5236 - val_acc: 0.7483
Epoch 30/100
390870/390870 [==============================] - 70s 180us/step - loss: 0.5165 - acc: 0.7512 - val_loss: 0.5249 - val_acc: 0.7474
Epoch 31/100
390870/390870 [==============================] - 68s 173us/step - loss: 0.5163 - acc: 0.7509 - val_loss: 0.5443 - val_acc: 0.7423
Epoch 32/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5157 - acc: 0.7513 - val_loss: 0.5229 - val_acc: 0.7493
Epoch 33/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5165 - acc: 0.7510 - val_loss: 0.5228 - val_acc: 0.7485
Epoch 34/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5151 - acc: 0.7518 - val_loss: 0.5211 - val_acc: 0.7493
Epoch 35/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5148 - acc: 0.7523 - val_loss: 0.5217 - val_acc: 0.7504
Epoch 36/100
390870/390870 [==============================] - 65s 165us/step - loss: 0.5142 - acc: 0.7519 - val_loss: 0.5242 - val_acc: 0.7489
Epoch 37/100
390870/390870 [==============================] - 65s 165us/step - loss: 0.5130 - acc: 0.7527 - val_loss: 0.5208 - val_acc: 0.7496
Epoch 38/100
390870/390870 [==============================] - 65s 165us/step - loss: 0.5132 - acc: 0.7533 - val_loss: 0.5242 - val_acc: 0.7489
Epoch 39/100
390870/390870 [==============================] - 65s 165us/step - loss: 0.5136 - acc: 0.7528 - val_loss: 0.5211 - val_acc: 0.7504
Epoch 40/100
390870/390870 [==============================] - 70s 179us/step - loss: 0.5131 - acc: 0.7529 - val_loss: 0.5211 - val_acc: 0.7520
Epoch 41/100
390870/390870 [==============================] - 69s 176us/step - loss: 0.5122 - acc: 0.7534 - val_loss: 0.5201 - val_acc: 0.7498
Epoch 42/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5125 - acc: 0.7531 - val_loss: 0.5371 - val_acc: 0.7479
Epoch 43/100
390870/390870 [==============================] - 67s 170us/step - loss: 0.5127 - acc: 0.7531 - val_loss: 0.5216 - val_acc: 0.7474
Epoch 44/100
390870/390870 [==============================] - 66s 170us/step - loss: 0.5124 - acc: 0.7535 - val_loss: 0.5237 - val_acc: 0.7483
Epoch 45/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5110 - acc: 0.7542 - val_loss: 0.5254 - val_acc: 0.7495
Epoch 46/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5116 - acc: 0.7541 - val_loss: 0.5196 - val_acc: 0.7490
Epoch 47/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5117 - acc: 0.7535 - val_loss: 0.5290 - val_acc: 0.7438
Epoch 48/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5112 - acc: 0.7543 - val_loss: 0.5260 - val_acc: 0.7486
Epoch 49/100
390870/390870 [==============================] - 65s 165us/step - loss: 0.5106 - acc: 0.7545 - val_loss: 0.5208 - val_acc: 0.7501
Epoch 50/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5106 - acc: 0.7540 - val_loss: 0.5226 - val_acc: 0.7493
Epoch 51/100
390870/390870 [==============================] - 63s 162us/step - loss: 0.5115 - acc: 0.7539 - val_loss: 0.5251 - val_acc: 0.7476
Epoch 52/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5101 - acc: 0.7547 - val_loss: 0.5229 - val_acc: 0.7481
Epoch 53/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5101 - acc: 0.7549 - val_loss: 0.5207 - val_acc: 0.7514
Epoch 54/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5099 - acc: 0.7551 - val_loss: 0.5192 - val_acc: 0.7514
Epoch 55/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5093 - acc: 0.7551 - val_loss: 0.5207 - val_acc: 0.7492
Epoch 56/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5095 - acc: 0.7552 - val_loss: 0.5232 - val_acc: 0.7467
Epoch 57/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5096 - acc: 0.7549 - val_loss: 0.5234 - val_acc: 0.7490
Epoch 58/100
390870/390870 [==============================] - 63s 162us/step - loss: 0.5091 - acc: 0.7550 - val_loss: 0.5194 - val_acc: 0.7495
Epoch 59/100
390870/390870 [==============================] - 63s 161us/step - loss: 0.5115 - acc: 0.7540 - val_loss: 0.5216 - val_acc: 0.7510
Epoch 60/100
390870/390870 [==============================] - 63s 161us/step - loss: 0.5085 - acc: 0.7554 - val_loss: 0.5181 - val_acc: 0.7511
Epoch 61/100
390870/390870 [==============================] - 63s 162us/step - loss: 0.5088 - acc: 0.7554 - val_loss: 0.5281 - val_acc: 0.7476
Epoch 62/100
390870/390870 [==============================] - 63s 162us/step - loss: 0.5089 - acc: 0.7552 - val_loss: 0.5264 - val_acc: 0.7459
Epoch 63/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5097 - acc: 0.7548 - val_loss: 0.5209 - val_acc: 0.7489
Epoch 64/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5105 - acc: 0.7545 - val_loss: 0.5215 - val_acc: 0.7495
Epoch 65/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5115 - acc: 0.7535 - val_loss: 0.5253 - val_acc: 0.7475
Epoch 66/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5101 - acc: 0.7554 - val_loss: 0.5272 - val_acc: 0.7430
Epoch 67/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5081 - acc: 0.7556 - val_loss: 0.5202 - val_acc: 0.7510
Epoch 68/100
390870/390870 [==============================] - 69s 176us/step - loss: 0.5076 - acc: 0.7558 - val_loss: 0.5277 - val_acc: 0.7463
Epoch 69/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5092 - acc: 0.7555 - val_loss: 0.5229 - val_acc: 0.7468
Epoch 70/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5090 - acc: 0.7557 - val_loss: 0.5229 - val_acc: 0.7491
Epoch 71/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5071 - acc: 0.7565 - val_loss: 0.5311 - val_acc: 0.7507
Epoch 72/100
390870/390870 [==============================] - 63s 161us/step - loss: 0.5076 - acc: 0.7561 - val_loss: 0.5203 - val_acc: 0.7509
Epoch 73/100
390870/390870 [==============================] - 63s 161us/step - loss: 0.5071 - acc: 0.7566 - val_loss: 0.5237 - val_acc: 0.7501
Epoch 74/100
390870/390870 [==============================] - 63s 161us/step - loss: 0.5090 - acc: 0.7554 - val_loss: 0.5276 - val_acc: 0.7463
Epoch 75/100
390870/390870 [==============================] - 63s 162us/step - loss: 0.5068 - acc: 0.7567 - val_loss: 0.5351 - val_acc: 0.7462
Epoch 76/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5062 - acc: 0.7570 - val_loss: 0.5302 - val_acc: 0.7477
Epoch 77/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5063 - acc: 0.7563 - val_loss: 0.5270 - val_acc: 0.7512
Epoch 78/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5060 - acc: 0.7569 - val_loss: 0.5287 - val_acc: 0.7438
Epoch 79/100
390870/390870 [==============================] - 64s 165us/step - loss: 0.5058 - acc: 0.7569 - val_loss: 0.5219 - val_acc: 0.7504
Epoch 80/100
390870/390870 [==============================] - 64s 165us/step - loss: 0.5059 - acc: 0.7569 - val_loss: 0.5373 - val_acc: 0.7476
Epoch 81/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5090 - acc: 0.7553 - val_loss: 0.5302 - val_acc: 0.7470
Epoch 82/100
390870/390870 [==============================] - 71s 182us/step - loss: 0.5064 - acc: 0.7566 - val_loss: 0.5293 - val_acc: 0.7498
Epoch 83/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5057 - acc: 0.7571 - val_loss: 0.5328 - val_acc: 0.7446
Epoch 84/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5052 - acc: 0.7575 - val_loss: 0.5275 - val_acc: 0.7510
Epoch 85/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5050 - acc: 0.7571 - val_loss: 0.5285 - val_acc: 0.7506
Epoch 86/100
390870/390870 [==============================] - 64s 162us/step - loss: 0.5054 - acc: 0.7573 - val_loss: 0.5313 - val_acc: 0.7499
Epoch 87/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5061 - acc: 0.7570 - val_loss: 0.5275 - val_acc: 0.7508
Epoch 88/100
390870/390870 [==============================] - 63s 161us/step - loss: 0.5086 - acc: 0.7554 - val_loss: 0.5396 - val_acc: 0.7454
Epoch 89/100
390870/390870 [==============================] - 63s 161us/step - loss: 0.5069 - acc: 0.7568 - val_loss: 0.5216 - val_acc: 0.7506
Epoch 90/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5083 - acc: 0.7551 - val_loss: 0.5246 - val_acc: 0.7483
Epoch 91/100
390870/390870 [==============================] - 65s 165us/step - loss: 0.5057 - acc: 0.7569 - val_loss: 0.5279 - val_acc: 0.7516
Epoch 92/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5062 - acc: 0.7564 - val_loss: 0.5280 - val_acc: 0.7495
Epoch 93/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5088 - acc: 0.7545 - val_loss: 0.5226 - val_acc: 0.7500
Epoch 94/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5078 - acc: 0.7561 - val_loss: 0.5331 - val_acc: 0.7493
Epoch 95/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5052 - acc: 0.7573 - val_loss: 0.5277 - val_acc: 0.7465
Epoch 96/100
390870/390870 [==============================] - 69s 177us/step - loss: 0.5046 - acc: 0.7574 - val_loss: 0.5355 - val_acc: 0.7459
Epoch 97/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5061 - acc: 0.7565 - val_loss: 0.5250 - val_acc: 0.7497
Epoch 98/100
390870/390870 [==============================] - 64s 165us/step - loss: 0.5034 - acc: 0.7584 - val_loss: 0.5233 - val_acc: 0.7513
Epoch 99/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5042 - acc: 0.7577 - val_loss: 0.5248 - val_acc: 0.7475
Epoch 100/100
390870/390870 [==============================] - 63s 161us/step - loss: 0.5036 - acc: 0.7579 - val_loss: 0.5342 - val_acc: 0.7499

<keras.callbacks.History at 0x1d81f160d30>

# Save trained model
classifier.save("../models/trained_deep_neural_network_all.h5")

# evaluation of the model training
evaluation = classifier.evaluate(train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])
print(evaluation)

558386/558386 [==============================] - 57s 101us/step
[0.512360177343886, 0.7563101510431093]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_2.pickle')
test_set.shape

(153993, 2774)

prediction = classifier.predict(test_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
binary_prediction = _binarize(prediction, threshold = 0.5)
binary_prediction

array([[1.],
       [0.],
       [1.],
       ...,
       [0.],
       [1.],
       [0.]], dtype=float32)

set(binary_prediction[:,0])

{0.0, 1.0}

print("Report for Deep Learning approach:")
print(_classification_report(test_set['likes'], binary_prediction))

print("Accuracy for Deep Learning approach:", _accuracy_score(test_set['likes'], binary_prediction) * 100)

Report for Deep Learning approach:
              precision    recall  f1-score   support

           0       0.71      0.38      0.49     50930
           1       0.75      0.92      0.83    103063

    accuracy                           0.74    153993
   macro avg       0.73      0.65      0.66    153993
weighted avg       0.74      0.74      0.72    153993

Accuracy for Deep Learning approach: 74.33649581474484

matrix = _confusion_matrix(test_set['likes'], binary_prediction)
print(matrix)

[[19270 31660]
 [ 7860 95203]]

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], binary_prediction)

_plt.plot(fpr, tpr)
_plt.xlim([0.0, 1.0])
_plt.ylim([0.0, 1.0])

_plt.title("Neural Network ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

NN 2¶

alpha = 2

# hidden layers = 3

batch size = 100

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_2.pickle')
train_set.shape

(558386, 2774)

# Number of neurons for hiddens layer
number_samples, number_features = train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']).shape

number_input_neurons = number_features
number_output_neurons = 1
number_train_semples = number_samples
alpha = 2

number_hidden_neurons = round(number_train_semples / (alpha * (number_input_neurons + number_output_neurons)))
print(number_hidden_neurons)

67

classifier = _Sequential()

# First Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal', input_dim = number_features))

# Second Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Third Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Output Layer
classifier.add(_Dense(1, activation = 'sigmoid', kernel_initializer = 'random_normal'))

# Compiling the neural network
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

# Fitting the data to the training dataset
classifier.fit(train_set.drop(
    columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']),
    train_set['likes'], validation_split = 0.3, batch_size = 100, epochs = 100)

Train on 390870 samples, validate on 167516 samples
Epoch 1/100
390870/390870 [==============================] - 76s 194us/step - loss: 0.5945 - acc: 0.6981 - val_loss: 0.5504 - val_acc: 0.7257
Epoch 2/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5560 - acc: 0.7232 - val_loss: 0.5542 - val_acc: 0.7265
Epoch 3/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5459 - acc: 0.7304 - val_loss: 0.5586 - val_acc: 0.7329
Epoch 4/100
390870/390870 [==============================] - 73s 186us/step - loss: 0.5414 - acc: 0.7336 - val_loss: 0.5431 - val_acc: 0.7366
Epoch 5/100
390870/390870 [==============================] - 69s 177us/step - loss: 0.5378 - acc: 0.7364 - val_loss: 0.5399 - val_acc: 0.7331
Epoch 6/100
390870/390870 [==============================] - 68s 174us/step - loss: 0.5355 - acc: 0.7382 - val_loss: 0.5342 - val_acc: 0.7421
Epoch 7/100
390870/390870 [==============================] - 68s 174us/step - loss: 0.5335 - acc: 0.7393 - val_loss: 0.5430 - val_acc: 0.7335
Epoch 8/100
390870/390870 [==============================] - 68s 173us/step - loss: 0.5329 - acc: 0.7400 - val_loss: 0.5299 - val_acc: 0.7448
Epoch 9/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5309 - acc: 0.7415 - val_loss: 0.5312 - val_acc: 0.7450
Epoch 10/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5290 - acc: 0.7422 - val_loss: 0.5402 - val_acc: 0.7360
Epoch 11/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5283 - acc: 0.7437 - val_loss: 0.5288 - val_acc: 0.7454
Epoch 12/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5258 - acc: 0.7447 - val_loss: 0.5303 - val_acc: 0.7431
Epoch 13/100
390870/390870 [==============================] - 65s 165us/step - loss: 0.5260 - acc: 0.7453 - val_loss: 0.5309 - val_acc: 0.7432
Epoch 14/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5247 - acc: 0.7456 - val_loss: 0.5333 - val_acc: 0.7417
Epoch 15/100
390870/390870 [==============================] - 65s 165us/step - loss: 0.5248 - acc: 0.7458 - val_loss: 0.5262 - val_acc: 0.7480
Epoch 16/100
390870/390870 [==============================] - 65s 165us/step - loss: 0.5227 - acc: 0.7471 - val_loss: 0.5845 - val_acc: 0.7252
Epoch 17/100
390870/390870 [==============================] - 65s 165us/step - loss: 0.5222 - acc: 0.7472 - val_loss: 0.5288 - val_acc: 0.7439
Epoch 18/100
390870/390870 [==============================] - 70s 180us/step - loss: 0.5222 - acc: 0.7472 - val_loss: 0.5242 - val_acc: 0.7481
Epoch 19/100
390870/390870 [==============================] - 69s 177us/step - loss: 0.5211 - acc: 0.7481 - val_loss: 0.5295 - val_acc: 0.7480
Epoch 20/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5202 - acc: 0.7478 - val_loss: 0.5281 - val_acc: 0.7466
Epoch 21/100
390870/390870 [==============================] - 68s 173us/step - loss: 0.5202 - acc: 0.7481 - val_loss: 0.5229 - val_acc: 0.7487
Epoch 22/100
390870/390870 [==============================] - 67s 170us/step - loss: 0.5186 - acc: 0.7495 - val_loss: 0.5255 - val_acc: 0.7470
Epoch 23/100
390870/390870 [==============================] - 68s 173us/step - loss: 0.5183 - acc: 0.7493 - val_loss: 0.5234 - val_acc: 0.7496
Epoch 24/100
390870/390870 [==============================] - 68s 173us/step - loss: 0.5175 - acc: 0.7502 - val_loss: 0.5376 - val_acc: 0.7437
Epoch 25/100
390870/390870 [==============================] - 69s 176us/step - loss: 0.5175 - acc: 0.7507 - val_loss: 0.5282 - val_acc: 0.7466
Epoch 26/100
390870/390870 [==============================] - 70s 179us/step - loss: 0.5170 - acc: 0.7500 - val_loss: 0.5223 - val_acc: 0.7488
Epoch 27/100
390870/390870 [==============================] - 70s 179us/step - loss: 0.5204 - acc: 0.7479 - val_loss: 0.5407 - val_acc: 0.7352
Epoch 28/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5160 - acc: 0.7510 - val_loss: 0.5215 - val_acc: 0.7495
Epoch 29/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5156 - acc: 0.7510 - val_loss: 0.5270 - val_acc: 0.7460
Epoch 30/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5145 - acc: 0.7519 - val_loss: 0.5264 - val_acc: 0.7464
Epoch 31/100
390870/390870 [==============================] - 72s 183us/step - loss: 0.5155 - acc: 0.7515 - val_loss: 0.5299 - val_acc: 0.7429
Epoch 32/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5154 - acc: 0.7516 - val_loss: 0.5260 - val_acc: 0.7458
Epoch 33/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5160 - acc: 0.7506 - val_loss: 0.5369 - val_acc: 0.7389
Epoch 34/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5151 - acc: 0.7512 - val_loss: 0.5578 - val_acc: 0.7236
Epoch 35/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5131 - acc: 0.7525 - val_loss: 0.5299 - val_acc: 0.7433
Epoch 36/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5125 - acc: 0.7526 - val_loss: 0.5241 - val_acc: 0.7469
Epoch 37/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5139 - acc: 0.7525 - val_loss: 0.5235 - val_acc: 0.7478
Epoch 38/100
390870/390870 [==============================] - 64s 165us/step - loss: 0.5135 - acc: 0.7529 - val_loss: 0.5226 - val_acc: 0.7495
Epoch 39/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5153 - acc: 0.7518 - val_loss: 0.5248 - val_acc: 0.7476
Epoch 40/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5110 - acc: 0.7542 - val_loss: 0.5360 - val_acc: 0.7444
Epoch 41/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5110 - acc: 0.7541 - val_loss: 0.5234 - val_acc: 0.7478
Epoch 42/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5109 - acc: 0.7545 - val_loss: 0.5260 - val_acc: 0.7481
Epoch 43/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5105 - acc: 0.7537 - val_loss: 0.5239 - val_acc: 0.7488
Epoch 44/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5109 - acc: 0.7540 - val_loss: 0.5303 - val_acc: 0.7405
Epoch 45/100
390870/390870 [==============================] - 70s 179us/step - loss: 0.5103 - acc: 0.7549 - val_loss: 0.5251 - val_acc: 0.7485
Epoch 46/100
390870/390870 [==============================] - 67s 170us/step - loss: 0.5092 - acc: 0.7550 - val_loss: 0.5380 - val_acc: 0.7445
Epoch 47/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5096 - acc: 0.7551 - val_loss: 0.5319 - val_acc: 0.7433
Epoch 48/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5082 - acc: 0.7555 - val_loss: 0.5193 - val_acc: 0.7508
Epoch 49/100
390870/390870 [==============================] - 65s 165us/step - loss: 0.5084 - acc: 0.7553 - val_loss: 0.5240 - val_acc: 0.7484
Epoch 50/100
390870/390870 [==============================] - 65s 165us/step - loss: 0.5092 - acc: 0.7554 - val_loss: 0.5313 - val_acc: 0.7443
Epoch 51/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5087 - acc: 0.7554 - val_loss: 0.5237 - val_acc: 0.7494
Epoch 52/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5074 - acc: 0.7562 - val_loss: 0.5261 - val_acc: 0.7464
Epoch 53/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5076 - acc: 0.7557 - val_loss: 0.5212 - val_acc: 0.7500
Epoch 54/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5068 - acc: 0.7558 - val_loss: 0.5263 - val_acc: 0.7494
Epoch 55/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5080 - acc: 0.7559 - val_loss: 0.5306 - val_acc: 0.7435
Epoch 56/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5069 - acc: 0.7564 - val_loss: 0.5259 - val_acc: 0.7492
Epoch 57/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5067 - acc: 0.7562 - val_loss: 0.5279 - val_acc: 0.7469
Epoch 58/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5069 - acc: 0.7559 - val_loss: 0.5317 - val_acc: 0.7491
Epoch 59/100
390870/390870 [==============================] - 70s 179us/step - loss: 0.5068 - acc: 0.7570 - val_loss: 0.5292 - val_acc: 0.7458
Epoch 60/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5057 - acc: 0.7573 - val_loss: 0.5221 - val_acc: 0.7499
Epoch 61/100
390870/390870 [==============================] - 65s 165us/step - loss: 0.5063 - acc: 0.7571 - val_loss: 0.5270 - val_acc: 0.7494
Epoch 62/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5055 - acc: 0.7571 - val_loss: 0.5219 - val_acc: 0.7496
Epoch 63/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5046 - acc: 0.7573 - val_loss: 0.5254 - val_acc: 0.7471
Epoch 64/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5055 - acc: 0.7571 - val_loss: 0.5264 - val_acc: 0.7494
Epoch 65/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5039 - acc: 0.7576 - val_loss: 0.5297 - val_acc: 0.7470
Epoch 66/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5045 - acc: 0.7574 - val_loss: 0.5283 - val_acc: 0.7481
Epoch 67/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5056 - acc: 0.7572 - val_loss: 0.5276 - val_acc: 0.7473
Epoch 68/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5059 - acc: 0.7566 - val_loss: 0.5221 - val_acc: 0.7494
Epoch 69/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5039 - acc: 0.7581 - val_loss: 0.5297 - val_acc: 0.7480
Epoch 70/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5033 - acc: 0.7584 - val_loss: 0.5300 - val_acc: 0.7451
Epoch 71/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5034 - acc: 0.7576 - val_loss: 0.5224 - val_acc: 0.7500
Epoch 72/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5023 - acc: 0.7587 - val_loss: 0.5275 - val_acc: 0.7484
Epoch 73/100
390870/390870 [==============================] - 71s 182us/step - loss: 0.5031 - acc: 0.7584 - val_loss: 0.5269 - val_acc: 0.7485
Epoch 74/100
390870/390870 [==============================] - 68s 174us/step - loss: 0.5027 - acc: 0.7584 - val_loss: 0.5260 - val_acc: 0.7488
Epoch 75/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5018 - acc: 0.7597 - val_loss: 0.5272 - val_acc: 0.7478
Epoch 76/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5026 - acc: 0.7590 - val_loss: 0.5292 - val_acc: 0.7493
Epoch 77/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5021 - acc: 0.7591 - val_loss: 0.5394 - val_acc: 0.7473
Epoch 78/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5029 - acc: 0.7586 - val_loss: 0.5319 - val_acc: 0.7476
Epoch 79/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5027 - acc: 0.7591 - val_loss: 0.5461 - val_acc: 0.7434
Epoch 80/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5021 - acc: 0.7592 - val_loss: 0.5273 - val_acc: 0.7498
Epoch 81/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5015 - acc: 0.7594 - val_loss: 0.5773 - val_acc: 0.7094
Epoch 82/100
390870/390870 [==============================] - 68s 173us/step - loss: 0.5009 - acc: 0.7600 - val_loss: 0.5287 - val_acc: 0.7478
Epoch 83/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5015 - acc: 0.7593 - val_loss: 0.5263 - val_acc: 0.7478
Epoch 84/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5016 - acc: 0.7588 - val_loss: 0.5305 - val_acc: 0.7472
Epoch 85/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5015 - acc: 0.7598 - val_loss: 0.5262 - val_acc: 0.7498
Epoch 86/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5003 - acc: 0.7598 - val_loss: 0.5268 - val_acc: 0.7487
Epoch 87/100
390870/390870 [==============================] - 71s 181us/step - loss: 0.5003 - acc: 0.7599 - val_loss: 0.5320 - val_acc: 0.7493
Epoch 88/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5043 - acc: 0.7576 - val_loss: 0.5263 - val_acc: 0.7477
Epoch 89/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5013 - acc: 0.7593 - val_loss: 0.5292 - val_acc: 0.7459
Epoch 90/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5003 - acc: 0.7599 - val_loss: 0.5309 - val_acc: 0.7497
Epoch 91/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5030 - acc: 0.7584 - val_loss: 0.5248 - val_acc: 0.7496
Epoch 92/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5012 - acc: 0.7596 - val_loss: 0.5330 - val_acc: 0.7478
Epoch 93/100
390870/390870 [==============================] - 65s 165us/step - loss: 0.5009 - acc: 0.7593 - val_loss: 0.5318 - val_acc: 0.7475
Epoch 94/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.4989 - acc: 0.7607 - val_loss: 0.5308 - val_acc: 0.7487
Epoch 95/100
390870/390870 [==============================] - 65s 168us/step - loss: 0.4990 - acc: 0.7607 - val_loss: 0.5308 - val_acc: 0.7497
Epoch 96/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.4990 - acc: 0.7613 - val_loss: 0.5352 - val_acc: 0.7477
Epoch 97/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.4985 - acc: 0.7607 - val_loss: 0.5291 - val_acc: 0.7488
Epoch 98/100
390870/390870 [==============================] - 65s 168us/step - loss: 0.4983 - acc: 0.7615 - val_loss: 0.5259 - val_acc: 0.7461
Epoch 99/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.4999 - acc: 0.7602 - val_loss: 0.5438 - val_acc: 0.7421
Epoch 100/100
390870/390870 [==============================] - 68s 174us/step - loss: 0.4978 - acc: 0.7614 - val_loss: 0.5337 - val_acc: 0.7468

<keras.callbacks.History at 0x1dcde2b5d68>

# Save trained model
classifier.save("../models/trained_deep_neural_network_all2.h5")

# evaluation of the model training
evaluation = classifier.evaluate(train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])
print(evaluation)

558386/558386 [==============================] - 51s 90us/step
[0.5102951304472344, 0.7552123441486025]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_2.pickle')
test_set.shape

(153993, 2774)

prediction = classifier.predict(test_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
binary_prediction = _binarize(prediction, threshold = 0.5)
binary_prediction

array([[1.],
       [0.],
       [1.],
       ...,
       [1.],
       [1.],
       [1.]], dtype=float32)

set(binary_prediction[:,0])

{0.0, 1.0}

print("Report for Deep Learning approach:")
print(_classification_report(test_set['likes'], binary_prediction))

print("Accuracy for Deep Learning approach:", _accuracy_score(test_set['likes'], binary_prediction) * 100)

Report for Deep Learning approach:
              precision    recall  f1-score   support

           0       0.73      0.33      0.46     50930
           1       0.74      0.94      0.83    103063

    accuracy                           0.74    153993
   macro avg       0.73      0.64      0.64    153993
weighted avg       0.74      0.74      0.70    153993

Accuracy for Deep Learning approach: 73.82478424343964

matrix = _confusion_matrix(test_set['likes'], binary_prediction)
print(matrix)

[[16881 34049]
 [ 6259 96804]]

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], binary_prediction)

_plt.plot(fpr, tpr)
_plt.xlim([0.0, 1.0])
_plt.ylim([0.0, 1.0])

_plt.title("Neural Network ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

NN 3¶

alpha = 7

# hidden layers = 3

batch size = 100

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_2.pickle')
train_set.shape

(558386, 2774)

# Number of neurons for hiddens layer
number_samples, number_features = train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']).shape

number_input_neurons = number_features
number_output_neurons = 1
number_train_semples = number_samples
alpha = 7

number_hidden_neurons = round(number_train_semples / (alpha * (number_input_neurons + number_output_neurons)))
print(number_hidden_neurons)

29

classifier = _Sequential()

# First Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal', input_dim = number_features))

# Second Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Third Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Output Layer
classifier.add(_Dense(1, activation = 'sigmoid', kernel_initializer = 'random_normal'))

# Compiling the neural network
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

# Fitting the data to the training dataset
classifier.fit(train_set.drop(
    columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']),
    train_set['likes'], validation_split = 0.3, batch_size = 100, epochs = 100)

Train on 390870 samples, validate on 167516 samples
Epoch 1/100
390870/390870 [==============================] - 82s 209us/step - loss: 0.5856 - acc: 0.7017 - val_loss: 0.5694 - val_acc: 0.7125
Epoch 2/100
390870/390870 [==============================] - 69s 176us/step - loss: 0.5533 - acc: 0.7251 - val_loss: 0.5492 - val_acc: 0.7293
Epoch 3/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5445 - acc: 0.7317 - val_loss: 0.5462 - val_acc: 0.7312
Epoch 4/100
390870/390870 [==============================] - 73s 187us/step - loss: 0.5410 - acc: 0.7339 - val_loss: 0.5389 - val_acc: 0.7389
Epoch 5/100
390870/390870 [==============================] - 68s 174us/step - loss: 0.5381 - acc: 0.7366 - val_loss: 0.5335 - val_acc: 0.7427
Epoch 6/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5354 - acc: 0.7376 - val_loss: 0.5365 - val_acc: 0.7390
Epoch 7/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5334 - acc: 0.7398 - val_loss: 0.5297 - val_acc: 0.7436
Epoch 8/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5324 - acc: 0.7405 - val_loss: 0.5356 - val_acc: 0.7383
Epoch 9/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5301 - acc: 0.7424 - val_loss: 0.5294 - val_acc: 0.7443
Epoch 10/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5293 - acc: 0.7425 - val_loss: 0.5266 - val_acc: 0.7462
Epoch 11/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5279 - acc: 0.7434 - val_loss: 0.5368 - val_acc: 0.7414
Epoch 12/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5273 - acc: 0.7445 - val_loss: 0.5278 - val_acc: 0.7459
Epoch 13/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5260 - acc: 0.7443 - val_loss: 0.5314 - val_acc: 0.7415
Epoch 14/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5261 - acc: 0.7449 - val_loss: 0.5342 - val_acc: 0.7440
Epoch 15/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5254 - acc: 0.7453 - val_loss: 0.5326 - val_acc: 0.7456
Epoch 16/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5241 - acc: 0.7460 - val_loss: 0.5312 - val_acc: 0.7421
Epoch 17/100
390870/390870 [==============================] - 67s 173us/step - loss: 0.5248 - acc: 0.7454 - val_loss: 0.5303 - val_acc: 0.7437
Epoch 18/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5236 - acc: 0.7460 - val_loss: 0.5257 - val_acc: 0.7455
Epoch 19/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5216 - acc: 0.7477 - val_loss: 0.5236 - val_acc: 0.7479
Epoch 20/100
390870/390870 [==============================] - 72s 184us/step - loss: 0.5218 - acc: 0.7474 - val_loss: 0.5258 - val_acc: 0.7471
Epoch 21/100
390870/390870 [==============================] - 73s 188us/step - loss: 0.5221 - acc: 0.7474 - val_loss: 0.5318 - val_acc: 0.7455
Epoch 22/100
390870/390870 [==============================] - 69s 176us/step - loss: 0.5209 - acc: 0.7478 - val_loss: 0.5259 - val_acc: 0.7465
Epoch 23/100
390870/390870 [==============================] - 69s 177us/step - loss: 0.5220 - acc: 0.7472 - val_loss: 0.5280 - val_acc: 0.7454
Epoch 24/100
390870/390870 [==============================] - 69s 177us/step - loss: 0.5215 - acc: 0.7482 - val_loss: 0.5226 - val_acc: 0.7485
Epoch 25/100
390870/390870 [==============================] - 66s 170us/step - loss: 0.5219 - acc: 0.7473 - val_loss: 0.5232 - val_acc: 0.7481
Epoch 26/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5193 - acc: 0.7486 - val_loss: 0.5268 - val_acc: 0.7462
Epoch 27/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5184 - acc: 0.7495 - val_loss: 0.5267 - val_acc: 0.7469
Epoch 28/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5189 - acc: 0.7495 - val_loss: 0.5243 - val_acc: 0.7468
Epoch 29/100
390870/390870 [==============================] - 64s 165us/step - loss: 0.5178 - acc: 0.7499 - val_loss: 0.5323 - val_acc: 0.7437
Epoch 30/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5177 - acc: 0.7498 - val_loss: 0.5235 - val_acc: 0.7464
Epoch 31/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5170 - acc: 0.7503 - val_loss: 0.5246 - val_acc: 0.7437
Epoch 32/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5164 - acc: 0.7512 - val_loss: 0.5275 - val_acc: 0.7441
Epoch 33/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5160 - acc: 0.7507 - val_loss: 0.5223 - val_acc: 0.7495
Epoch 34/100
390870/390870 [==============================] - 70s 180us/step - loss: 0.5156 - acc: 0.7514 - val_loss: 0.5273 - val_acc: 0.7480
Epoch 35/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5156 - acc: 0.7514 - val_loss: 0.5261 - val_acc: 0.7478
Epoch 36/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5147 - acc: 0.7515 - val_loss: 0.5208 - val_acc: 0.7484
Epoch 37/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5148 - acc: 0.7517 - val_loss: 0.5228 - val_acc: 0.7475
Epoch 38/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5145 - acc: 0.7511 - val_loss: 0.5331 - val_acc: 0.7420
Epoch 39/100
390870/390870 [==============================] - 64s 165us/step - loss: 0.5140 - acc: 0.7521 - val_loss: 0.5244 - val_acc: 0.7469
Epoch 40/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5134 - acc: 0.7525 - val_loss: 0.5236 - val_acc: 0.7451
Epoch 41/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5137 - acc: 0.7521 - val_loss: 0.5223 - val_acc: 0.7479
Epoch 42/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5142 - acc: 0.7516 - val_loss: 0.5231 - val_acc: 0.7477
Epoch 43/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5129 - acc: 0.7530 - val_loss: 0.5260 - val_acc: 0.7479
Epoch 44/100
390870/390870 [==============================] - 64s 165us/step - loss: 0.5140 - acc: 0.7520 - val_loss: 0.5233 - val_acc: 0.7477
Epoch 45/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5131 - acc: 0.7522 - val_loss: 0.5250 - val_acc: 0.7470
Epoch 46/100
390870/390870 [==============================] - 65s 165us/step - loss: 0.5142 - acc: 0.7520 - val_loss: 0.5305 - val_acc: 0.7462
Epoch 47/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5134 - acc: 0.7525 - val_loss: 0.5262 - val_acc: 0.7454
Epoch 48/100
390870/390870 [==============================] - 69s 177us/step - loss: 0.5118 - acc: 0.7535 - val_loss: 0.5210 - val_acc: 0.7506
Epoch 49/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5119 - acc: 0.7536 - val_loss: 0.5244 - val_acc: 0.7473
Epoch 50/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5121 - acc: 0.7537 - val_loss: 0.5229 - val_acc: 0.7498
Epoch 51/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5118 - acc: 0.7534 - val_loss: 0.5242 - val_acc: 0.7471
Epoch 52/100
390870/390870 [==============================] - 65s 165us/step - loss: 0.5109 - acc: 0.7543 - val_loss: 0.5290 - val_acc: 0.7461
Epoch 53/100
390870/390870 [==============================] - 64s 165us/step - loss: 0.5110 - acc: 0.7537 - val_loss: 0.5242 - val_acc: 0.7479
Epoch 54/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5114 - acc: 0.7533 - val_loss: 0.5283 - val_acc: 0.7450
Epoch 55/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5109 - acc: 0.7540 - val_loss: 0.5282 - val_acc: 0.7471
Epoch 56/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5105 - acc: 0.7545 - val_loss: 0.5238 - val_acc: 0.7469
Epoch 57/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5102 - acc: 0.7547 - val_loss: 0.5273 - val_acc: 0.7467
Epoch 58/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5102 - acc: 0.7545 - val_loss: 0.5283 - val_acc: 0.7459
Epoch 59/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5103 - acc: 0.7540 - val_loss: 0.5335 - val_acc: 0.7429
Epoch 60/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5104 - acc: 0.7539 - val_loss: 0.5281 - val_acc: 0.7426
Epoch 61/100
390870/390870 [==============================] - 64s 165us/step - loss: 0.5099 - acc: 0.7544 - val_loss: 0.5273 - val_acc: 0.7482
Epoch 62/100
390870/390870 [==============================] - 69s 178us/step - loss: 0.5114 - acc: 0.7540 - val_loss: 0.5351 - val_acc: 0.7420
Epoch 63/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5107 - acc: 0.7540 - val_loss: 0.5387 - val_acc: 0.7308
Epoch 64/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5099 - acc: 0.7547 - val_loss: 0.5230 - val_acc: 0.7477
Epoch 65/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5098 - acc: 0.7548 - val_loss: 0.5240 - val_acc: 0.7484
Epoch 66/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5092 - acc: 0.7552 - val_loss: 0.5368 - val_acc: 0.7424
Epoch 67/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5087 - acc: 0.7551 - val_loss: 0.5328 - val_acc: 0.7424
Epoch 68/100
390870/390870 [==============================] - 63s 162us/step - loss: 0.5097 - acc: 0.7542 - val_loss: 0.5264 - val_acc: 0.7458
Epoch 69/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5085 - acc: 0.7557 - val_loss: 0.5245 - val_acc: 0.7476
Epoch 70/100
390870/390870 [==============================] - 66s 170us/step - loss: 0.5086 - acc: 0.7556 - val_loss: 0.5275 - val_acc: 0.7456
Epoch 71/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5086 - acc: 0.7547 - val_loss: 0.5261 - val_acc: 0.7472
Epoch 72/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5085 - acc: 0.7556 - val_loss: 0.5221 - val_acc: 0.7496
Epoch 73/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5084 - acc: 0.7551 - val_loss: 0.5211 - val_acc: 0.7501
Epoch 74/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5088 - acc: 0.7554 - val_loss: 0.5263 - val_acc: 0.7473
Epoch 75/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5091 - acc: 0.7546 - val_loss: 0.5258 - val_acc: 0.7480
Epoch 76/100
390870/390870 [==============================] - 72s 185us/step - loss: 0.5073 - acc: 0.7561 - val_loss: 0.5252 - val_acc: 0.7488
Epoch 77/100
390870/390870 [==============================] - 68s 175us/step - loss: 0.5082 - acc: 0.7550 - val_loss: 0.5250 - val_acc: 0.7473
Epoch 78/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5072 - acc: 0.7559 - val_loss: 0.5376 - val_acc: 0.7452
Epoch 79/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5082 - acc: 0.7559 - val_loss: 0.5283 - val_acc: 0.7422
Epoch 80/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5081 - acc: 0.7551 - val_loss: 0.5314 - val_acc: 0.7409
Epoch 81/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5068 - acc: 0.7567 - val_loss: 0.5225 - val_acc: 0.7490
Epoch 82/100
390870/390870 [==============================] - 64s 165us/step - loss: 0.5067 - acc: 0.7562 - val_loss: 0.5258 - val_acc: 0.7491
Epoch 83/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5080 - acc: 0.7559 - val_loss: 0.5203 - val_acc: 0.7497
Epoch 84/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5062 - acc: 0.7567 - val_loss: 0.5245 - val_acc: 0.7490
Epoch 85/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5069 - acc: 0.7559 - val_loss: 0.5207 - val_acc: 0.7505
Epoch 86/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5067 - acc: 0.7566 - val_loss: 0.5253 - val_acc: 0.7466
Epoch 87/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5066 - acc: 0.7565 - val_loss: 0.5364 - val_acc: 0.7421
Epoch 88/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5062 - acc: 0.7562 - val_loss: 0.5209 - val_acc: 0.7493
Epoch 89/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5061 - acc: 0.7562 - val_loss: 0.5227 - val_acc: 0.7481
Epoch 90/100
390870/390870 [==============================] - 70s 179us/step - loss: 0.5058 - acc: 0.7566 - val_loss: 0.5244 - val_acc: 0.7472
Epoch 91/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5061 - acc: 0.7562 - val_loss: 0.5256 - val_acc: 0.7465
Epoch 92/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5057 - acc: 0.7563 - val_loss: 0.5331 - val_acc: 0.7471
Epoch 93/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5055 - acc: 0.7568 - val_loss: 0.5243 - val_acc: 0.7490
Epoch 94/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5056 - acc: 0.7565 - val_loss: 0.5261 - val_acc: 0.7478
Epoch 95/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5058 - acc: 0.7570 - val_loss: 0.5244 - val_acc: 0.7467
Epoch 96/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5049 - acc: 0.7573 - val_loss: 0.5215 - val_acc: 0.7513
Epoch 97/100
390870/390870 [==============================] - 64s 164us/step - loss: 0.5062 - acc: 0.7567 - val_loss: 0.5249 - val_acc: 0.7499
Epoch 98/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5066 - acc: 0.7565 - val_loss: 0.5254 - val_acc: 0.7505
Epoch 99/100
390870/390870 [==============================] - 64s 165us/step - loss: 0.5047 - acc: 0.7572 - val_loss: 0.5241 - val_acc: 0.7479
Epoch 100/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5046 - acc: 0.7572 - val_loss: 0.5229 - val_acc: 0.7483

<keras.callbacks.History at 0x1e2ac885a58>

# Save trained model
classifier.save("../models/trained_deep_neural_network_all3.h5")

# evaluation of the model training
evaluation = classifier.evaluate(train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])
print(evaluation)

558386/558386 [==============================] - 54s 96us/step
[0.5092164321641403, 0.7539193317881909]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_2.pickle')
test_set.shape

(153993, 2774)

prediction = classifier.predict(test_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
binary_prediction = _binarize(prediction, threshold = 0.5)
binary_prediction

array([[1.],
       [0.],
       [1.],
       ...,
       [0.],
       [1.],
       [0.]], dtype=float32)

set(binary_prediction[:,0])

{0.0, 1.0}

print("Report for Deep Learning approach:")
print(_classification_report(test_set['likes'], binary_prediction))

print("Accuracy for Deep Learning approach:", _accuracy_score(test_set['likes'], binary_prediction) * 100)

Report for Deep Learning approach:
              precision    recall  f1-score   support

           0       0.70      0.38      0.49     50930
           1       0.75      0.92      0.83    103063

    accuracy                           0.74    153993
   macro avg       0.73      0.65      0.66    153993
weighted avg       0.74      0.74      0.72    153993

Accuracy for Deep Learning approach: 74.23843940958356

matrix = _confusion_matrix(test_set['likes'], binary_prediction)
print(matrix)

[[19398 31532]
 [ 8139 94924]]

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], binary_prediction)

_plt.plot(fpr, tpr)
_plt.xlim([0.0, 1.0])
_plt.ylim([0.0, 1.0])

_plt.title("Neural Network ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_del_all()

NN 4¶

alpha = 7

# hidden layers = 5

batch size = 100

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_2.pickle')
train_set.shape

(558386, 2774)

# Number of neurons for hiddens layer
number_samples, number_features = train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']).shape

number_input_neurons = number_features
number_output_neurons = 1
number_train_semples = number_samples
alpha = 7

number_hidden_neurons = round(number_train_semples / (alpha * (number_input_neurons + number_output_neurons)))
print(number_hidden_neurons)

29

classifier = _Sequential()

# First Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal', input_dim = number_features))

# Second Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Third Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fourth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fifth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Output Layer
classifier.add(_Dense(1, activation = 'sigmoid', kernel_initializer = 'random_normal'))

# Compiling the neural network
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

# Fitting the data to the training dataset
classifier.fit(train_set.drop(
    columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']),
    train_set['likes'], validation_split = 0.3, batch_size = 100, epochs = 100)

Train on 390870 samples, validate on 167516 samples
Epoch 1/100
390870/390870 [==============================] - 76s 194us/step - loss: 0.5860 - acc: 0.7003 - val_loss: 0.5558 - val_acc: 0.7246
Epoch 2/100
390870/390870 [==============================] - 68s 175us/step - loss: 0.5507 - acc: 0.7276 - val_loss: 0.5472 - val_acc: 0.7286
Epoch 3/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5437 - acc: 0.7324 - val_loss: 0.5416 - val_acc: 0.7332
Epoch 4/100
390870/390870 [==============================] - 67s 170us/step - loss: 0.5389 - acc: 0.7367 - val_loss: 0.5404 - val_acc: 0.7350
Epoch 5/100
390870/390870 [==============================] - 70s 180us/step - loss: 0.5355 - acc: 0.7380 - val_loss: 0.5455 - val_acc: 0.7335
Epoch 6/100
390870/390870 [==============================] - 82s 210us/step - loss: 0.5338 - acc: 0.7389 - val_loss: 0.5305 - val_acc: 0.7453
Epoch 7/100
390870/390870 [==============================] - 70s 179us/step - loss: 0.5318 - acc: 0.7409 - val_loss: 0.5410 - val_acc: 0.7380
Epoch 8/100
390870/390870 [==============================] - 69s 178us/step - loss: 0.5310 - acc: 0.7413 - val_loss: 0.5284 - val_acc: 0.7453
Epoch 9/100
390870/390870 [==============================] - 70s 178us/step - loss: 0.5294 - acc: 0.7424 - val_loss: 0.5343 - val_acc: 0.7369
Epoch 10/100
390870/390870 [==============================] - 69s 178us/step - loss: 0.5278 - acc: 0.7440 - val_loss: 0.5295 - val_acc: 0.7431
Epoch 11/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5272 - acc: 0.7438 - val_loss: 0.5369 - val_acc: 0.7398
Epoch 12/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5273 - acc: 0.7443 - val_loss: 0.5292 - val_acc: 0.7453
Epoch 13/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5255 - acc: 0.7451 - val_loss: 0.5319 - val_acc: 0.7404
Epoch 14/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5257 - acc: 0.7452 - val_loss: 0.5338 - val_acc: 0.7419
Epoch 15/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5247 - acc: 0.7460 - val_loss: 0.5309 - val_acc: 0.7463
Epoch 16/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5230 - acc: 0.7470 - val_loss: 0.5235 - val_acc: 0.7487
Epoch 17/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5225 - acc: 0.7472 - val_loss: 0.5274 - val_acc: 0.7448
Epoch 18/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5218 - acc: 0.7477 - val_loss: 0.5264 - val_acc: 0.7470
Epoch 19/100
390870/390870 [==============================] - 67s 170us/step - loss: 0.5210 - acc: 0.7483 - val_loss: 0.5228 - val_acc: 0.7489
Epoch 20/100
390870/390870 [==============================] - 74s 189us/step - loss: 0.5203 - acc: 0.7481 - val_loss: 0.5470 - val_acc: 0.7400
Epoch 21/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5203 - acc: 0.7486 - val_loss: 0.5242 - val_acc: 0.7466
Epoch 22/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5194 - acc: 0.7491 - val_loss: 0.5245 - val_acc: 0.7475
Epoch 23/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5192 - acc: 0.7496 - val_loss: 0.5237 - val_acc: 0.7484
Epoch 24/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5187 - acc: 0.7499 - val_loss: 0.5245 - val_acc: 0.7465
Epoch 25/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5189 - acc: 0.7495 - val_loss: 0.5257 - val_acc: 0.7472
Epoch 26/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5177 - acc: 0.7501 - val_loss: 0.5225 - val_acc: 0.7483
Epoch 27/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5175 - acc: 0.7505 - val_loss: 0.5334 - val_acc: 0.7459
Epoch 28/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5164 - acc: 0.7506 - val_loss: 0.5292 - val_acc: 0.7443
Epoch 29/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5173 - acc: 0.7504 - val_loss: 0.5240 - val_acc: 0.7487
Epoch 30/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5161 - acc: 0.7515 - val_loss: 0.5209 - val_acc: 0.7504
Epoch 31/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5160 - acc: 0.7512 - val_loss: 0.5322 - val_acc: 0.7455
Epoch 32/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5155 - acc: 0.7513 - val_loss: 0.5241 - val_acc: 0.7481
Epoch 33/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5147 - acc: 0.7520 - val_loss: 0.5242 - val_acc: 0.7483
Epoch 34/100
390870/390870 [==============================] - 72s 184us/step - loss: 0.5151 - acc: 0.7522 - val_loss: 0.5218 - val_acc: 0.7489
Epoch 35/100
390870/390870 [==============================] - 68s 175us/step - loss: 0.5145 - acc: 0.7526 - val_loss: 0.5315 - val_acc: 0.7455
Epoch 36/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5145 - acc: 0.7524 - val_loss: 0.5201 - val_acc: 0.7499
Epoch 37/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5143 - acc: 0.7519 - val_loss: 0.5216 - val_acc: 0.7484
Epoch 38/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5135 - acc: 0.7530 - val_loss: 0.5245 - val_acc: 0.7464
Epoch 39/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5133 - acc: 0.7526 - val_loss: 0.5236 - val_acc: 0.7473
Epoch 40/100
390870/390870 [==============================] - 68s 174us/step - loss: 0.5127 - acc: 0.7528 - val_loss: 0.5242 - val_acc: 0.7474
Epoch 41/100
390870/390870 [==============================] - 68s 173us/step - loss: 0.5129 - acc: 0.7524 - val_loss: 0.5211 - val_acc: 0.7483
Epoch 42/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5122 - acc: 0.7535 - val_loss: 0.5401 - val_acc: 0.7440
Epoch 43/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5121 - acc: 0.7537 - val_loss: 0.5244 - val_acc: 0.7481
Epoch 44/100
390870/390870 [==============================] - 68s 174us/step - loss: 0.5139 - acc: 0.7533 - val_loss: 0.5190 - val_acc: 0.7515
Epoch 45/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5117 - acc: 0.7536 - val_loss: 0.5238 - val_acc: 0.7503
Epoch 46/100
390870/390870 [==============================] - 68s 173us/step - loss: 0.5113 - acc: 0.7544 - val_loss: 0.5190 - val_acc: 0.7514
Epoch 47/100
390870/390870 [==============================] - 74s 188us/step - loss: 0.5112 - acc: 0.7541 - val_loss: 0.5202 - val_acc: 0.7507
Epoch 48/100
390870/390870 [==============================] - 68s 174us/step - loss: 0.5108 - acc: 0.7543 - val_loss: 0.5216 - val_acc: 0.7516
Epoch 49/100
390870/390870 [==============================] - 72s 183us/step - loss: 0.5107 - acc: 0.7543 - val_loss: 0.5286 - val_acc: 0.7482
Epoch 50/100
390870/390870 [==============================] - 70s 180us/step - loss: 0.5116 - acc: 0.7535 - val_loss: 0.5198 - val_acc: 0.7505
Epoch 51/100
390870/390870 [==============================] - 68s 173us/step - loss: 0.5111 - acc: 0.7538 - val_loss: 0.5203 - val_acc: 0.7493
Epoch 52/100
390870/390870 [==============================] - 71s 182us/step - loss: 0.5107 - acc: 0.7546 - val_loss: 0.5203 - val_acc: 0.7514ss: 0.5107 - acc: 
Epoch 53/100
390870/390870 [==============================] - 71s 183us/step - loss: 0.5107 - acc: 0.7540 - val_loss: 0.5216 - val_acc: 0.7491
Epoch 54/100
390870/390870 [==============================] - 70s 179us/step - loss: 0.5097 - acc: 0.7548 - val_loss: 0.5263 - val_acc: 0.7467
Epoch 55/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5091 - acc: 0.7549 - val_loss: 0.5264 - val_acc: 0.7455
Epoch 56/100
390870/390870 [==============================] - 70s 178us/step - loss: 0.5092 - acc: 0.7555 - val_loss: 0.5273 - val_acc: 0.7493
Epoch 57/100
390870/390870 [==============================] - 70s 179us/step - loss: 0.5091 - acc: 0.7556 - val_loss: 0.5221 - val_acc: 0.7493
Epoch 58/100
390870/390870 [==============================] - 70s 178us/step - loss: 0.5085 - acc: 0.7552 - val_loss: 0.5192 - val_acc: 0.7517
Epoch 59/100
390870/390870 [==============================] - 70s 178us/step - loss: 0.5089 - acc: 0.7554 - val_loss: 0.5267 - val_acc: 0.7464
Epoch 60/100
390870/390870 [==============================] - 75s 193us/step - loss: 0.5088 - acc: 0.7550 - val_loss: 0.5227 - val_acc: 0.7498
Epoch 61/100
390870/390870 [==============================] - 71s 181us/step - loss: 0.5082 - acc: 0.7555 - val_loss: 0.5191 - val_acc: 0.7517
Epoch 62/100
390870/390870 [==============================] - 67s 173us/step - loss: 0.5086 - acc: 0.7555 - val_loss: 0.5218 - val_acc: 0.7511
Epoch 63/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5088 - acc: 0.7551 - val_loss: 0.5211 - val_acc: 0.7500
Epoch 64/100
390870/390870 [==============================] - 68s 173us/step - loss: 0.5085 - acc: 0.7554 - val_loss: 0.5192 - val_acc: 0.7514
Epoch 65/100
390870/390870 [==============================] - 72s 184us/step - loss: 0.5078 - acc: 0.7552 - val_loss: 0.5199 - val_acc: 0.7503
Epoch 66/100
390870/390870 [==============================] - 70s 180us/step - loss: 0.5080 - acc: 0.7558 - val_loss: 0.5218 - val_acc: 0.7495
Epoch 67/100
390870/390870 [==============================] - 74s 189us/step - loss: 0.5084 - acc: 0.7552 - val_loss: 0.5231 - val_acc: 0.7485
Epoch 68/100
390870/390870 [==============================] - 70s 178us/step - loss: 0.5076 - acc: 0.7558 - val_loss: 0.5306 - val_acc: 0.7455
Epoch 69/100
390870/390870 [==============================] - 73s 187us/step - loss: 0.5073 - acc: 0.7560 - val_loss: 0.5270 - val_acc: 0.7504
Epoch 70/100
390870/390870 [==============================] - 73s 187us/step - loss: 0.5071 - acc: 0.7564 - val_loss: 0.5233 - val_acc: 0.7502
Epoch 71/100
390870/390870 [==============================] - 73s 186us/step - loss: 0.5076 - acc: 0.7558 - val_loss: 0.5326 - val_acc: 0.7471
Epoch 72/100
390870/390870 [==============================] - 69s 176us/step - loss: 0.5070 - acc: 0.7566 - val_loss: 0.5188 - val_acc: 0.7518
Epoch 73/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5071 - acc: 0.7562 - val_loss: 0.5230 - val_acc: 0.7499
Epoch 74/100
390870/390870 [==============================] - 70s 178us/step - loss: 0.5064 - acc: 0.7566 - val_loss: 0.5304 - val_acc: 0.7472
Epoch 75/100
390870/390870 [==============================] - 74s 190us/step - loss: 0.5064 - acc: 0.7567 - val_loss: 0.5190 - val_acc: 0.7519
Epoch 76/100
390870/390870 [==============================] - 70s 180us/step - loss: 0.5066 - acc: 0.7568 - val_loss: 0.5248 - val_acc: 0.7470
Epoch 77/100
390870/390870 [==============================] - 69s 176us/step - loss: 0.5066 - acc: 0.7563 - val_loss: 0.5291 - val_acc: 0.7492
Epoch 78/100
390870/390870 [==============================] - 69s 176us/step - loss: 0.5059 - acc: 0.7571 - val_loss: 0.5242 - val_acc: 0.7491
Epoch 79/100
390870/390870 [==============================] - 69s 178us/step - loss: 0.5058 - acc: 0.7569 - val_loss: 0.5234 - val_acc: 0.7493
Epoch 80/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5057 - acc: 0.7571 - val_loss: 0.5217 - val_acc: 0.7515
Epoch 81/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5058 - acc: 0.7571 - val_loss: 0.5193 - val_acc: 0.7515
Epoch 82/100
390870/390870 [==============================] - 67s 170us/step - loss: 0.5056 - acc: 0.7568 - val_loss: 0.5207 - val_acc: 0.7497
Epoch 83/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5052 - acc: 0.7577 - val_loss: 0.5324 - val_acc: 0.7428
Epoch 84/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5056 - acc: 0.7568 - val_loss: 0.5264 - val_acc: 0.7496
Epoch 85/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5056 - acc: 0.7573 - val_loss: 0.5225 - val_acc: 0.7511
Epoch 86/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5046 - acc: 0.7576 - val_loss: 0.5194 - val_acc: 0.7522
Epoch 87/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5052 - acc: 0.7569 - val_loss: 0.5229 - val_acc: 0.7509
Epoch 88/100
390870/390870 [==============================] - 72s 184us/step - loss: 0.5056 - acc: 0.7569 - val_loss: 0.5284 - val_acc: 0.7495
Epoch 89/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5052 - acc: 0.7573 - val_loss: 0.5270 - val_acc: 0.7508
Epoch 90/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5052 - acc: 0.7574 - val_loss: 0.5216 - val_acc: 0.7515
Epoch 91/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5047 - acc: 0.7582 - val_loss: 0.5327 - val_acc: 0.7487
Epoch 92/100
390870/390870 [==============================] - 66s 170us/step - loss: 0.5042 - acc: 0.7577 - val_loss: 0.5241 - val_acc: 0.7491
Epoch 93/100
390870/390870 [==============================] - 98s 251us/step - loss: 0.5041 - acc: 0.7578 - val_loss: 0.5283 - val_acc: 0.7463
Epoch 94/100
390870/390870 [==============================] - 152s 389us/step - loss: 0.5042 - acc: 0.7578 - val_loss: 0.5267 - val_acc: 0.7499
Epoch 95/100
390870/390870 [==============================] - 173s 442us/step - loss: 0.5038 - acc: 0.7579 - val_loss: 0.5319 - val_acc: 0.7487
Epoch 96/100
390870/390870 [==============================] - 67s 170us/step - loss: 0.5040 - acc: 0.7580 - val_loss: 0.5238 - val_acc: 0.7511
Epoch 97/100
390870/390870 [==============================] - 67s 170us/step - loss: 0.5039 - acc: 0.7580 - val_loss: 0.5258 - val_acc: 0.7498
Epoch 98/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5035 - acc: 0.7583 - val_loss: 0.5252 - val_acc: 0.7506
Epoch 99/100
390870/390870 [==============================] - 73s 188us/step - loss: 0.5043 - acc: 0.7576 - val_loss: 0.5237 - val_acc: 0.7501
Epoch 100/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5038 - acc: 0.7580 - val_loss: 0.5238 - val_acc: 0.7511

<keras.callbacks.History at 0x1e2e3f38630>

# Save trained model
classifier.save("../models/trained_deep_neural_network_all5.h5")

# evaluation of the model training
evaluation = classifier.evaluate(train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])
print(evaluation)

558386/558386 [==============================] - 53s 95us/step
[0.5069767637241182, 0.757692707194954]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_2.pickle')
test_set.shape

(153993, 2774)

prediction = classifier.predict(test_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
binary_prediction = _binarize(prediction, threshold = 0.5)
binary_prediction

array([[1.],
       [0.],
       [1.],
       ...,
       [0.],
       [1.],
       [0.]], dtype=float32)

set(binary_prediction[:,0])

{0.0, 1.0}

print("Report for Deep Learning approach:")
print(_classification_report(test_set['likes'], binary_prediction))

print("Accuracy for Deep Learning approach:", _accuracy_score(test_set['likes'], binary_prediction) * 100)

Report for Deep Learning approach:
              precision    recall  f1-score   support

           0       0.69      0.43      0.53     50930
           1       0.76      0.90      0.83    103063

    accuracy                           0.75    153993
   macro avg       0.72      0.67      0.68    153993
weighted avg       0.74      0.75      0.73    153993

Accuracy for Deep Learning approach: 74.6254699888956

matrix = _confusion_matrix(test_set['likes'], binary_prediction)
print(matrix)

[[21761 29169]
 [ 9906 93157]]

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], binary_prediction)

_plt.plot(fpr, tpr)
_plt.xlim([0.0, 1.0])
_plt.ylim([0.0, 1.0])

_plt.title("Neural Network ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_del_all()

NN 5¶

alpha = 7

# hidden layers = 5

batch size = 500

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_2.pickle')
train_set.shape

(558386, 2774)

# Number of neurons for hiddens layer
number_samples, number_features = train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']).shape

number_input_neurons = number_features
number_output_neurons = 1
number_train_semples = number_samples
alpha = 7

number_hidden_neurons = round(number_train_semples / (alpha * (number_input_neurons + number_output_neurons)))
print(number_hidden_neurons)

29

classifier = _Sequential()

# First Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal', input_dim = number_features))

# Second Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Third Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fourth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fifth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Output Layer
classifier.add(_Dense(1, activation = 'sigmoid', kernel_initializer = 'random_normal'))

# Compiling the neural network
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

# Fitting the data to the training dataset
classifier.fit(train_set.drop(
    columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']),
    train_set['likes'], validation_split = 0.3, batch_size = 500, epochs = 100)

Train on 390870 samples, validate on 167516 samples
Epoch 1/100
390870/390870 [==============================] - 57s 145us/step - loss: 0.6050 - acc: 0.6886 - val_loss: 0.5680 - val_acc: 0.7174
Epoch 2/100
390870/390870 [==============================] - 54s 138us/step - loss: 0.5585 - acc: 0.7214 - val_loss: 0.5550 - val_acc: 0.7256
Epoch 3/100
390870/390870 [==============================] - 53s 136us/step - loss: 0.5464 - acc: 0.7309 - val_loss: 0.5400 - val_acc: 0.7387
Epoch 4/100
390870/390870 [==============================] - 53s 135us/step - loss: 0.5406 - acc: 0.7345 - val_loss: 0.5605 - val_acc: 0.7164
Epoch 5/100
390870/390870 [==============================] - 53s 135us/step - loss: 0.5363 - acc: 0.7370 - val_loss: 0.5421 - val_acc: 0.7324
Epoch 6/100
390870/390870 [==============================] - 54s 137us/step - loss: 0.5341 - acc: 0.7390 - val_loss: 0.5412 - val_acc: 0.7340
Epoch 7/100
390870/390870 [==============================] - 53s 136us/step - loss: 0.5315 - acc: 0.7410 - val_loss: 0.5354 - val_acc: 0.7407
Epoch 8/100
390870/390870 [==============================] - 58s 149us/step - loss: 0.5304 - acc: 0.7415 - val_loss: 0.5401 - val_acc: 0.7375
Epoch 9/100
390870/390870 [==============================] - 54s 138us/step - loss: 0.5303 - acc: 0.7414 - val_loss: 0.5321 - val_acc: 0.7445
Epoch 10/100
390870/390870 [==============================] - 53s 134us/step - loss: 0.5277 - acc: 0.7431 - val_loss: 0.5394 - val_acc: 0.7354
Epoch 11/100
390870/390870 [==============================] - 53s 135us/step - loss: 0.5261 - acc: 0.7448 - val_loss: 0.5291 - val_acc: 0.7442
Epoch 12/100
390870/390870 [==============================] - 54s 139us/step - loss: 0.5260 - acc: 0.7445 - val_loss: 0.5308 - val_acc: 0.7437
Epoch 13/100
390870/390870 [==============================] - 53s 136us/step - loss: 0.5248 - acc: 0.7453 - val_loss: 0.5329 - val_acc: 0.7410
Epoch 14/100
390870/390870 [==============================] - 53s 135us/step - loss: 0.5240 - acc: 0.7463 - val_loss: 0.5294 - val_acc: 0.7449
Epoch 15/100
390870/390870 [==============================] - 53s 135us/step - loss: 0.5226 - acc: 0.7468 - val_loss: 0.5345 - val_acc: 0.7397
Epoch 16/100
390870/390870 [==============================] - 53s 135us/step - loss: 0.5220 - acc: 0.7478 - val_loss: 0.5270 - val_acc: 0.7451
Epoch 17/100
390870/390870 [==============================] - 54s 139us/step - loss: 0.5216 - acc: 0.7480 - val_loss: 0.5347 - val_acc: 0.7375
Epoch 18/100
390870/390870 [==============================] - 53s 135us/step - loss: 0.5216 - acc: 0.7479 - val_loss: 0.5338 - val_acc: 0.7427
Epoch 19/100
390870/390870 [==============================] - 53s 135us/step - loss: 0.5198 - acc: 0.7488 - val_loss: 0.5268 - val_acc: 0.7446
Epoch 20/100
390870/390870 [==============================] - 53s 136us/step - loss: 0.5201 - acc: 0.7490 - val_loss: 0.5260 - val_acc: 0.7464
Epoch 21/100
390870/390870 [==============================] - 53s 135us/step - loss: 0.5199 - acc: 0.7491 - val_loss: 0.5272 - val_acc: 0.7472
Epoch 22/100
390870/390870 [==============================] - 53s 135us/step - loss: 0.5187 - acc: 0.7503 - val_loss: 0.5312 - val_acc: 0.7460
Epoch 23/100
390870/390870 [==============================] - 53s 135us/step - loss: 0.5182 - acc: 0.7498 - val_loss: 0.5255 - val_acc: 0.7470
Epoch 24/100
390870/390870 [==============================] - 53s 135us/step - loss: 0.5185 - acc: 0.7500 - val_loss: 0.5231 - val_acc: 0.7478
Epoch 25/100
390870/390870 [==============================] - 55s 142us/step - loss: 0.5164 - acc: 0.7515 - val_loss: 0.5231 - val_acc: 0.7482
Epoch 26/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5180 - acc: 0.7499 - val_loss: 0.5375 - val_acc: 0.7416
Epoch 27/100
390870/390870 [==============================] - 52s 134us/step - loss: 0.5158 - acc: 0.7514 - val_loss: 0.5224 - val_acc: 0.7485
Epoch 28/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5167 - acc: 0.7513 - val_loss: 0.5280 - val_acc: 0.7449
Epoch 29/100
390870/390870 [==============================] - 52s 134us/step - loss: 0.5155 - acc: 0.7518 - val_loss: 0.5221 - val_acc: 0.7489
Epoch 30/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5144 - acc: 0.7524 - val_loss: 0.5238 - val_acc: 0.7487
Epoch 31/100
390870/390870 [==============================] - 52s 134us/step - loss: 0.5143 - acc: 0.7524 - val_loss: 0.5236 - val_acc: 0.7475
Epoch 32/100
390870/390870 [==============================] - 52s 134us/step - loss: 0.5152 - acc: 0.7516 - val_loss: 0.5229 - val_acc: 0.7478
Epoch 33/100
390870/390870 [==============================] - 52s 134us/step - loss: 0.5146 - acc: 0.7525 - val_loss: 0.5210 - val_acc: 0.7502
Epoch 34/100
390870/390870 [==============================] - 52s 134us/step - loss: 0.5137 - acc: 0.7525 - val_loss: 0.5263 - val_acc: 0.7480
Epoch 35/100
390870/390870 [==============================] - 52s 134us/step - loss: 0.5141 - acc: 0.7520 - val_loss: 0.5218 - val_acc: 0.7504
Epoch 36/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5126 - acc: 0.7537 - val_loss: 0.5221 - val_acc: 0.7487
Epoch 37/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5128 - acc: 0.7533 - val_loss: 0.5234 - val_acc: 0.7476
Epoch 38/100
390870/390870 [==============================] - 51s 130us/step - loss: 0.5128 - acc: 0.7529 - val_loss: 0.5282 - val_acc: 0.7457
Epoch 39/100
390870/390870 [==============================] - 51s 130us/step - loss: 0.5127 - acc: 0.7532 - val_loss: 0.5275 - val_acc: 0.7450
Epoch 40/100
390870/390870 [==============================] - 51s 131us/step - loss: 0.5123 - acc: 0.7532 - val_loss: 0.5284 - val_acc: 0.7454
Epoch 41/100
390870/390870 [==============================] - 51s 130us/step - loss: 0.5106 - acc: 0.7540 - val_loss: 0.5223 - val_acc: 0.7506
Epoch 42/100
390870/390870 [==============================] - 53s 136us/step - loss: 0.5110 - acc: 0.7539 - val_loss: 0.5356 - val_acc: 0.7429
Epoch 43/100
390870/390870 [==============================] - 53s 135us/step - loss: 0.5106 - acc: 0.7546 - val_loss: 0.5221 - val_acc: 0.7499
Epoch 44/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5114 - acc: 0.7540 - val_loss: 0.5223 - val_acc: 0.7497
Epoch 45/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5099 - acc: 0.7547 - val_loss: 0.5279 - val_acc: 0.7487
Epoch 46/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5097 - acc: 0.7551 - val_loss: 0.5227 - val_acc: 0.7501
Epoch 47/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5096 - acc: 0.7554 - val_loss: 0.5270 - val_acc: 0.7467
Epoch 48/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5095 - acc: 0.7552 - val_loss: 0.5200 - val_acc: 0.7490
Epoch 49/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5096 - acc: 0.7549 - val_loss: 0.5226 - val_acc: 0.7484
Epoch 50/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5095 - acc: 0.7551 - val_loss: 0.5250 - val_acc: 0.7493
Epoch 51/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5080 - acc: 0.7559 - val_loss: 0.5202 - val_acc: 0.7504
Epoch 52/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5088 - acc: 0.7552 - val_loss: 0.5212 - val_acc: 0.7499
Epoch 53/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5083 - acc: 0.7554 - val_loss: 0.5253 - val_acc: 0.7473
Epoch 54/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5082 - acc: 0.7559 - val_loss: 0.5205 - val_acc: 0.7509
Epoch 55/100
390870/390870 [==============================] - 51s 131us/step - loss: 0.5081 - acc: 0.7554 - val_loss: 0.5253 - val_acc: 0.7487
Epoch 56/100
390870/390870 [==============================] - 51s 129us/step - loss: 0.5080 - acc: 0.7555 - val_loss: 0.5207 - val_acc: 0.7509
Epoch 57/100
390870/390870 [==============================] - 51s 130us/step - loss: 0.5069 - acc: 0.7568 - val_loss: 0.5214 - val_acc: 0.7497
Epoch 58/100
390870/390870 [==============================] - 51s 131us/step - loss: 0.5073 - acc: 0.7564 - val_loss: 0.5199 - val_acc: 0.7509
Epoch 59/100
390870/390870 [==============================] - 77s 198us/step - loss: 0.5079 - acc: 0.7558 - val_loss: 0.5295 - val_acc: 0.7464
Epoch 60/100
390870/390870 [==============================] - 54s 138us/step - loss: 0.5064 - acc: 0.7564 - val_loss: 0.5291 - val_acc: 0.7469
Epoch 61/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5065 - acc: 0.7568 - val_loss: 0.5185 - val_acc: 0.7520
Epoch 62/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5065 - acc: 0.7568 - val_loss: 0.5221 - val_acc: 0.7496
Epoch 63/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5059 - acc: 0.7571 - val_loss: 0.5276 - val_acc: 0.7488
Epoch 64/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5057 - acc: 0.7574 - val_loss: 0.5204 - val_acc: 0.7505
Epoch 65/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5055 - acc: 0.7577 - val_loss: 0.5293 - val_acc: 0.7427
Epoch 66/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5046 - acc: 0.7582 - val_loss: 0.5190 - val_acc: 0.7508
Epoch 67/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5052 - acc: 0.7575 - val_loss: 0.5196 - val_acc: 0.7517
Epoch 68/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5049 - acc: 0.7580 - val_loss: 0.5246 - val_acc: 0.7480
Epoch 69/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5054 - acc: 0.7577 - val_loss: 0.5297 - val_acc: 0.7441
Epoch 70/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5044 - acc: 0.7576 - val_loss: 0.5272 - val_acc: 0.7490
Epoch 71/100
390870/390870 [==============================] - 52s 132us/step - loss: 0.5058 - acc: 0.7574 - val_loss: 0.5225 - val_acc: 0.7481
Epoch 72/100
390870/390870 [==============================] - 51s 130us/step - loss: 0.5039 - acc: 0.7583 - val_loss: 0.5221 - val_acc: 0.7486
Epoch 73/100
390870/390870 [==============================] - 60s 153us/step - loss: 0.5041 - acc: 0.7584 - val_loss: 0.5193 - val_acc: 0.7506
Epoch 74/100
390870/390870 [==============================] - 51s 130us/step - loss: 0.5035 - acc: 0.7582 - val_loss: 0.5223 - val_acc: 0.7491
Epoch 75/100
390870/390870 [==============================] - 51s 129us/step - loss: 0.5033 - acc: 0.7588 - val_loss: 0.5227 - val_acc: 0.7506
Epoch 76/100
390870/390870 [==============================] - 51s 129us/step - loss: 0.5033 - acc: 0.7582 - val_loss: 0.5282 - val_acc: 0.7470
Epoch 77/100
390870/390870 [==============================] - 55s 141us/step - loss: 0.5042 - acc: 0.7580 - val_loss: 0.5244 - val_acc: 0.7499
Epoch 78/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5034 - acc: 0.7586 - val_loss: 0.5261 - val_acc: 0.7466
Epoch 79/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5027 - acc: 0.7595 - val_loss: 0.5263 - val_acc: 0.7494
Epoch 80/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5028 - acc: 0.7592 - val_loss: 0.5273 - val_acc: 0.7460
Epoch 81/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5022 - acc: 0.7593 - val_loss: 0.5257 - val_acc: 0.7488
Epoch 82/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5025 - acc: 0.7589 - val_loss: 0.5220 - val_acc: 0.7502
Epoch 83/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5019 - acc: 0.7593 - val_loss: 0.5270 - val_acc: 0.7478
Epoch 84/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5023 - acc: 0.7594 - val_loss: 0.5259 - val_acc: 0.7492
Epoch 85/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5018 - acc: 0.7592 - val_loss: 0.5267 - val_acc: 0.7498
Epoch 86/100
390870/390870 [==============================] - 52s 134us/step - loss: 0.5021 - acc: 0.7591 - val_loss: 0.5302 - val_acc: 0.7445
Epoch 87/100
390870/390870 [==============================] - 52s 134us/step - loss: 0.5013 - acc: 0.7597 - val_loss: 0.5209 - val_acc: 0.7512
Epoch 88/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5010 - acc: 0.7596 - val_loss: 0.5268 - val_acc: 0.7466
Epoch 89/100
390870/390870 [==============================] - 51s 131us/step - loss: 0.5019 - acc: 0.7594 - val_loss: 0.5236 - val_acc: 0.7510
Epoch 90/100
390870/390870 [==============================] - 51s 129us/step - loss: 0.5012 - acc: 0.7598 - val_loss: 0.5240 - val_acc: 0.7489
Epoch 91/100
390870/390870 [==============================] - 51s 130us/step - loss: 0.5016 - acc: 0.7598 - val_loss: 0.5236 - val_acc: 0.7516
Epoch 92/100
390870/390870 [==============================] - 51s 130us/step - loss: 0.5013 - acc: 0.7596 - val_loss: 0.5237 - val_acc: 0.7501
Epoch 93/100
390870/390870 [==============================] - 51s 130us/step - loss: 0.5009 - acc: 0.7597 - val_loss: 0.5250 - val_acc: 0.7505
Epoch 94/100
390870/390870 [==============================] - 55s 142us/step - loss: 0.5003 - acc: 0.7599 - val_loss: 0.5223 - val_acc: 0.7501
Epoch 95/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5002 - acc: 0.7598 - val_loss: 0.5274 - val_acc: 0.7493
Epoch 96/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5002 - acc: 0.7599 - val_loss: 0.5247 - val_acc: 0.7497
Epoch 97/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5000 - acc: 0.7601 - val_loss: 0.5289 - val_acc: 0.7498
Epoch 98/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.4997 - acc: 0.7612 - val_loss: 0.5273 - val_acc: 0.7504
Epoch 99/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.4997 - acc: 0.7605 - val_loss: 0.5269 - val_acc: 0.7495
Epoch 100/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.4994 - acc: 0.7607 - val_loss: 0.5307 - val_acc: 0.7475

<keras.callbacks.History at 0x28588237f28>

# Save trained model
classifier.save("../models/trained_deep_neural_network_all6.h5")

# evaluation of the model training
evaluation = classifier.evaluate(train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])
print(evaluation)

558386/558386 [==============================] - 46s 83us/step
[0.5109344036041277, 0.7546177733688839]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_2.pickle')
test_set.shape

(153993, 2774)

prediction = classifier.predict(test_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
binary_prediction = _binarize(prediction, threshold = 0.5)
binary_prediction

array([[1.],
       [0.],
       [1.],
       ...,
       [1.],
       [1.],
       [1.]], dtype=float32)

set(binary_prediction[:,0])

{0.0, 1.0}

print("Report for Deep Learning approach:")
print(_classification_report(test_set['likes'], binary_prediction))

print("Accuracy for Deep Learning approach:", _accuracy_score(test_set['likes'], binary_prediction) * 100)

Report for Deep Learning approach:
              precision    recall  f1-score   support

           0       0.73      0.34      0.46     50930
           1       0.74      0.94      0.83    103063

    accuracy                           0.74    153993
   macro avg       0.74      0.64      0.65    153993
weighted avg       0.74      0.74      0.71    153993

Accuracy for Deep Learning approach: 73.97803796276455

matrix = _confusion_matrix(test_set['likes'], binary_prediction)
print(matrix)

[[17292 33638]
 [ 6434 96629]]

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], binary_prediction)

_plt.plot(fpr, tpr)
_plt.xlim([0.0, 1.0])
_plt.ylim([0.0, 1.0])

_plt.title("Neural Network ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_del_all()

NN 6¶

alpha = 6

# hidden layers = 5

batch size = 100

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_2.pickle')
train_set.shape

(558386, 2774)

# Number of neurons for hiddens layer
number_samples, number_features = train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']).shape

number_input_neurons = number_features
number_output_neurons = 1
number_train_semples = number_samples
alpha = 6

number_hidden_neurons = round(number_train_semples / (alpha * (number_input_neurons + number_output_neurons)))
print(number_hidden_neurons)

34

classifier = _Sequential()

# First Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal', input_dim = number_features))

# Second Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Third Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fourth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fifth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Output Layer
classifier.add(_Dense(1, activation = 'sigmoid', kernel_initializer = 'random_normal'))

# Compiling the neural network
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

# Fitting the data to the training dataset
classifier.fit(train_set.drop(
    columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']),
    train_set['likes'], validation_split = 0.3, batch_size = 100, epochs = 100)

Train on 390870 samples, validate on 167516 samples
Epoch 1/100
390870/390870 [==============================] - 71s 183us/step - loss: 0.5860 - acc: 0.7012 - val_loss: 0.5567 - val_acc: 0.7211
Epoch 2/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5531 - acc: 0.7259 - val_loss: 0.5461 - val_acc: 0.7325
Epoch 3/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5453 - acc: 0.7313 - val_loss: 0.5470 - val_acc: 0.7292
Epoch 4/100
390870/390870 [==============================] - 65s 167us/step - loss: 0.5406 - acc: 0.7342 - val_loss: 0.5556 - val_acc: 0.7176
Epoch 5/100
390870/390870 [==============================] - 65s 166us/step - loss: 0.5377 - acc: 0.7368 - val_loss: 0.5365 - val_acc: 0.7417
Epoch 6/100
390870/390870 [==============================] - 70s 179us/step - loss: 0.5350 - acc: 0.7385 - val_loss: 0.5380 - val_acc: 0.7371
Epoch 7/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5324 - acc: 0.7404 - val_loss: 0.5296 - val_acc: 0.7453
Epoch 8/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5307 - acc: 0.7412 - val_loss: 0.5569 - val_acc: 0.7224
Epoch 9/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5291 - acc: 0.7430 - val_loss: 0.5283 - val_acc: 0.7449
Epoch 10/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5279 - acc: 0.7433 - val_loss: 0.5364 - val_acc: 0.7405
Epoch 11/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5261 - acc: 0.7454 - val_loss: 0.5308 - val_acc: 0.7451
Epoch 12/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5257 - acc: 0.7454 - val_loss: 0.5291 - val_acc: 0.7446
Epoch 13/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5246 - acc: 0.7458 - val_loss: 0.5282 - val_acc: 0.7453
Epoch 14/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5236 - acc: 0.7467 - val_loss: 0.5402 - val_acc: 0.7408
Epoch 15/100
390870/390870 [==============================] - 66s 170us/step - loss: 0.5226 - acc: 0.7469 - val_loss: 0.5656 - val_acc: 0.7069
Epoch 16/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5218 - acc: 0.7479 - val_loss: 0.5318 - val_acc: 0.7428
Epoch 17/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5214 - acc: 0.7483 - val_loss: 0.5306 - val_acc: 0.7426
Epoch 18/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5205 - acc: 0.7484 - val_loss: 0.5240 - val_acc: 0.7477
Epoch 19/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5203 - acc: 0.7486 - val_loss: 0.5261 - val_acc: 0.7462
Epoch 20/100
390870/390870 [==============================] - 72s 183us/step - loss: 0.5192 - acc: 0.7494 - val_loss: 0.5250 - val_acc: 0.7462
Epoch 21/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5190 - acc: 0.7495 - val_loss: 0.5293 - val_acc: 0.7428
Epoch 22/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5178 - acc: 0.7504 - val_loss: 0.5246 - val_acc: 0.7484
Epoch 23/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5172 - acc: 0.7507 - val_loss: 0.5313 - val_acc: 0.7447
Epoch 24/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5168 - acc: 0.7506 - val_loss: 0.5272 - val_acc: 0.7495
Epoch 25/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5168 - acc: 0.7504 - val_loss: 0.5352 - val_acc: 0.7368
Epoch 26/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5159 - acc: 0.7516 - val_loss: 0.5378 - val_acc: 0.7412
Epoch 27/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5163 - acc: 0.7510 - val_loss: 0.5219 - val_acc: 0.7488
Epoch 28/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5150 - acc: 0.7519 - val_loss: 0.5265 - val_acc: 0.7484
Epoch 29/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5142 - acc: 0.7520 - val_loss: 0.5203 - val_acc: 0.7489
Epoch 30/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5146 - acc: 0.7520 - val_loss: 0.5282 - val_acc: 0.7444
Epoch 31/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5138 - acc: 0.7522 - val_loss: 0.5228 - val_acc: 0.7481
Epoch 32/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5147 - acc: 0.7522 - val_loss: 0.5244 - val_acc: 0.7488
Epoch 33/100
390870/390870 [==============================] - 71s 181us/step - loss: 0.5133 - acc: 0.7528 - val_loss: 0.5283 - val_acc: 0.7472
Epoch 34/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5128 - acc: 0.7530 - val_loss: 0.5404 - val_acc: 0.7392
Epoch 35/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5131 - acc: 0.7530 - val_loss: 0.5231 - val_acc: 0.7475
Epoch 36/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5124 - acc: 0.7530 - val_loss: 0.5216 - val_acc: 0.7472
Epoch 37/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5121 - acc: 0.7531 - val_loss: 0.5203 - val_acc: 0.7502
Epoch 38/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5122 - acc: 0.7538 - val_loss: 0.5307 - val_acc: 0.7488
Epoch 39/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5108 - acc: 0.7542 - val_loss: 0.5241 - val_acc: 0.7482
Epoch 40/100
390870/390870 [==============================] - 68s 173us/step - loss: 0.5113 - acc: 0.7533 - val_loss: 0.5227 - val_acc: 0.7492
Epoch 41/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5105 - acc: 0.7546 - val_loss: 0.5185 - val_acc: 0.7502
Epoch 42/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5106 - acc: 0.7542 - val_loss: 0.5242 - val_acc: 0.7478
Epoch 43/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5099 - acc: 0.7543 - val_loss: 0.5295 - val_acc: 0.7486
Epoch 44/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5097 - acc: 0.7545 - val_loss: 0.5188 - val_acc: 0.7514
Epoch 45/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5098 - acc: 0.7547 - val_loss: 0.5262 - val_acc: 0.7463
Epoch 46/100
390870/390870 [==============================] - 66s 170us/step - loss: 0.5092 - acc: 0.7556 - val_loss: 0.5207 - val_acc: 0.7503
Epoch 47/100
390870/390870 [==============================] - 72s 183us/step - loss: 0.5089 - acc: 0.7551 - val_loss: 0.5212 - val_acc: 0.7496
Epoch 48/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5089 - acc: 0.7551 - val_loss: 0.5202 - val_acc: 0.7500
Epoch 49/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5085 - acc: 0.7552 - val_loss: 0.5279 - val_acc: 0.7454
Epoch 50/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5083 - acc: 0.7557 - val_loss: 0.5201 - val_acc: 0.7491
Epoch 51/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5079 - acc: 0.7560 - val_loss: 0.5370 - val_acc: 0.7446
Epoch 52/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5075 - acc: 0.7559 - val_loss: 0.5251 - val_acc: 0.7470
Epoch 53/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5079 - acc: 0.7562 - val_loss: 0.5235 - val_acc: 0.7495
Epoch 54/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5073 - acc: 0.7567 - val_loss: 0.5264 - val_acc: 0.7460
Epoch 55/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5070 - acc: 0.7565 - val_loss: 0.5272 - val_acc: 0.7467
Epoch 56/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5071 - acc: 0.7565 - val_loss: 0.5214 - val_acc: 0.7506
Epoch 57/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5068 - acc: 0.7563 - val_loss: 0.5289 - val_acc: 0.7469
Epoch 58/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5070 - acc: 0.7561 - val_loss: 0.5300 - val_acc: 0.7478
Epoch 59/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5062 - acc: 0.7567 - val_loss: 0.5801 - val_acc: 0.7259
Epoch 60/100
390870/390870 [==============================] - 70s 180us/step - loss: 0.5056 - acc: 0.7570 - val_loss: 0.5222 - val_acc: 0.7502
Epoch 61/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5059 - acc: 0.7568 - val_loss: 0.5229 - val_acc: 0.7499
Epoch 62/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5056 - acc: 0.7569 - val_loss: 0.5210 - val_acc: 0.7501
Epoch 63/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5055 - acc: 0.7574 - val_loss: 0.5225 - val_acc: 0.7503
Epoch 64/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5059 - acc: 0.7569 - val_loss: 0.5217 - val_acc: 0.7521
Epoch 65/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5050 - acc: 0.7572 - val_loss: 0.5291 - val_acc: 0.7444
Epoch 66/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5047 - acc: 0.7576 - val_loss: 0.5266 - val_acc: 0.7461
Epoch 67/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5047 - acc: 0.7576 - val_loss: 0.5246 - val_acc: 0.7491
Epoch 68/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5047 - acc: 0.7573 - val_loss: 0.5227 - val_acc: 0.7500
Epoch 69/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5041 - acc: 0.7581 - val_loss: 0.5244 - val_acc: 0.7497
Epoch 70/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5041 - acc: 0.7578 - val_loss: 0.5185 - val_acc: 0.7517
Epoch 71/100
390870/390870 [==============================] - 67s 170us/step - loss: 0.5045 - acc: 0.7575 - val_loss: 0.5247 - val_acc: 0.7488
Epoch 72/100
390870/390870 [==============================] - 66s 168us/step - loss: 0.5039 - acc: 0.7584 - val_loss: 0.5315 - val_acc: 0.7481
Epoch 73/100
390870/390870 [==============================] - 67s 170us/step - loss: 0.5038 - acc: 0.7583 - val_loss: 0.5238 - val_acc: 0.7498
Epoch 74/100
390870/390870 [==============================] - 72s 185us/step - loss: 0.5034 - acc: 0.7580 - val_loss: 0.5238 - val_acc: 0.7497
Epoch 75/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5033 - acc: 0.7587 - val_loss: 0.5217 - val_acc: 0.7495
Epoch 76/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5028 - acc: 0.7582 - val_loss: 0.5268 - val_acc: 0.7469
Epoch 77/100
390870/390870 [==============================] - 67s 170us/step - loss: 0.5033 - acc: 0.7584 - val_loss: 0.5227 - val_acc: 0.7494
Epoch 78/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5028 - acc: 0.7586 - val_loss: 0.5263 - val_acc: 0.7477
Epoch 79/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5022 - acc: 0.7589 - val_loss: 0.5261 - val_acc: 0.7477
Epoch 80/100
390870/390870 [==============================] - 67s 170us/step - loss: 0.5023 - acc: 0.7590 - val_loss: 0.5231 - val_acc: 0.7491
Epoch 81/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5022 - acc: 0.7592 - val_loss: 0.5223 - val_acc: 0.7512
Epoch 82/100
390870/390870 [==============================] - 66s 170us/step - loss: 0.5018 - acc: 0.7593 - val_loss: 0.5234 - val_acc: 0.7496
Epoch 83/100
390870/390870 [==============================] - 173s 443us/step - loss: 0.5016 - acc: 0.7592 - val_loss: 0.5249 - val_acc: 0.7528
Epoch 84/100
390870/390870 [==============================] - 69s 176us/step - loss: 0.5019 - acc: 0.7590 - val_loss: 0.5272 - val_acc: 0.7499
Epoch 85/100
390870/390870 [==============================] - 69s 177us/step - loss: 0.5018 - acc: 0.7591 - val_loss: 0.5320 - val_acc: 0.7462
Epoch 86/100
390870/390870 [==============================] - 75s 193us/step - loss: 0.5014 - acc: 0.7593 - val_loss: 0.5244 - val_acc: 0.7509
Epoch 87/100
390870/390870 [==============================] - 71s 181us/step - loss: 0.5011 - acc: 0.7597 - val_loss: 0.5241 - val_acc: 0.7508
Epoch 88/100
390870/390870 [==============================] - 69s 177us/step - loss: 0.5008 - acc: 0.7597 - val_loss: 0.5280 - val_acc: 0.7500
Epoch 89/100
390870/390870 [==============================] - 69s 177us/step - loss: 0.5007 - acc: 0.7595 - val_loss: 0.5217 - val_acc: 0.7519
Epoch 90/100
390870/390870 [==============================] - 69s 177us/step - loss: 0.5014 - acc: 0.7594 - val_loss: 0.5226 - val_acc: 0.7497
Epoch 91/100
390870/390870 [==============================] - 68s 174us/step - loss: 0.5009 - acc: 0.7596 - val_loss: 0.5245 - val_acc: 0.7496
Epoch 92/100
390870/390870 [==============================] - 67s 173us/step - loss: 0.5010 - acc: 0.7590 - val_loss: 0.5284 - val_acc: 0.7489
Epoch 93/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.5005 - acc: 0.7598 - val_loss: 0.5343 - val_acc: 0.7456
Epoch 94/100
390870/390870 [==============================] - 67s 171us/step - loss: 0.5003 - acc: 0.7597 - val_loss: 0.5248 - val_acc: 0.7487
Epoch 95/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5003 - acc: 0.7601 - val_loss: 0.5264 - val_acc: 0.7486
Epoch 96/100
390870/390870 [==============================] - 66s 170us/step - loss: 0.4998 - acc: 0.7605 - val_loss: 0.5259 - val_acc: 0.7515
Epoch 97/100
390870/390870 [==============================] - 66s 169us/step - loss: 0.5000 - acc: 0.7607 - val_loss: 0.5246 - val_acc: 0.7491
Epoch 98/100
390870/390870 [==============================] - 66s 170us/step - loss: 0.4996 - acc: 0.7610 - val_loss: 0.5377 - val_acc: 0.7410
Epoch 99/100
390870/390870 [==============================] - 72s 184us/step - loss: 0.5007 - acc: 0.7601 - val_loss: 0.5272 - val_acc: 0.7484
Epoch 100/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.4996 - acc: 0.7608 - val_loss: 0.5210 - val_acc: 0.7509

<keras.callbacks.History at 0x28588c76ac8>

# Save trained model
classifier.save("../models/trained_deep_neural_network_all7.h5")

# evaluation of the model training
evaluation = classifier.evaluate(train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])
print(evaluation)

558386/558386 [==============================] - 48s 87us/step
[0.5033024442936678, 0.7597557961701017]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_2.pickle')
test_set.shape

(153993, 2774)

prediction = classifier.predict(test_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
binary_prediction = _binarize(prediction, threshold = 0.5)
binary_prediction

array([[1.],
       [0.],
       [1.],
       ...,
       [1.],
       [1.],
       [0.]], dtype=float32)

set(binary_prediction[:,0])

{0.0, 1.0}

print("Report for Deep Learning approach:")
print(_classification_report(test_set['likes'], binary_prediction))

print("Accuracy for Deep Learning approach:", _accuracy_score(test_set['likes'], binary_prediction) * 100)

Report for Deep Learning approach:
              precision    recall  f1-score   support

           0       0.69      0.42      0.52     50930
           1       0.76      0.91      0.83    103063

    accuracy                           0.74    153993
   macro avg       0.72      0.66      0.67    153993
weighted avg       0.74      0.74      0.72    153993

Accuracy for Deep Learning approach: 74.48585325306995

matrix = _confusion_matrix(test_set['likes'], binary_prediction)
print(matrix)

[[21192 29738]
 [ 9552 93511]]

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], binary_prediction)

_plt.plot(fpr, tpr)
_plt.xlim([0.0, 1.0])
_plt.ylim([0.0, 1.0])

_plt.title("Neural Network ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_del_all()

NN 7¶

alpha = 6

# hidden layers = 5

batch size = 500

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_2.pickle')
train_set.shape

(558386, 2774)

# Number of neurons for hiddens layer
number_samples, number_features = train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']).shape

number_input_neurons = number_features
number_output_neurons = 1
number_train_semples = number_samples
alpha = 6

number_hidden_neurons = round(number_train_semples / (alpha * (number_input_neurons + number_output_neurons)))
print(number_hidden_neurons)

34

classifier = _Sequential()

# First Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal', input_dim = number_features))

# Second Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Third Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fourth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fifth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Output Layer
classifier.add(_Dense(1, activation = 'sigmoid', kernel_initializer = 'random_normal'))

# Compiling the neural network
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

# Fitting the data to the training dataset
classifier.fit(train_set.drop(
    columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']),
    train_set['likes'], validation_split = 0.3, batch_size = 500, epochs = 100)

Train on 390870 samples, validate on 167516 samples
Epoch 1/100
390870/390870 [==============================] - 60s 153us/step - loss: 0.6000 - acc: 0.6893 - val_loss: 0.5623 - val_acc: 0.7238
Epoch 2/100
390870/390870 [==============================] - 52s 134us/step - loss: 0.5547 - acc: 0.7240 - val_loss: 0.5505 - val_acc: 0.7272
Epoch 3/100
390870/390870 [==============================] - 51s 131us/step - loss: 0.5446 - acc: 0.7316 - val_loss: 0.5395 - val_acc: 0.7382
Epoch 4/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5374 - acc: 0.7368 - val_loss: 0.5323 - val_acc: 0.7432
Epoch 5/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5366 - acc: 0.7371 - val_loss: 0.5371 - val_acc: 0.7398
Epoch 6/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5339 - acc: 0.7391 - val_loss: 0.5352 - val_acc: 0.7395
Epoch 7/100
390870/390870 [==============================] - 51s 130us/step - loss: 0.5312 - acc: 0.7411 - val_loss: 0.5340 - val_acc: 0.7425
Epoch 8/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5290 - acc: 0.7423 - val_loss: 0.5493 - val_acc: 0.7272
Epoch 9/100
390870/390870 [==============================] - 52s 134us/step - loss: 0.5284 - acc: 0.7431 - val_loss: 0.5344 - val_acc: 0.7406
Epoch 10/100
390870/390870 [==============================] - 57s 145us/step - loss: 0.5262 - acc: 0.7445 - val_loss: 0.5320 - val_acc: 0.7417
Epoch 11/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5248 - acc: 0.7456 - val_loss: 0.5352 - val_acc: 0.7399
Epoch 12/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5244 - acc: 0.7457 - val_loss: 0.5249 - val_acc: 0.7468
Epoch 13/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5227 - acc: 0.7468 - val_loss: 0.5284 - val_acc: 0.7451
Epoch 14/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5223 - acc: 0.7481 - val_loss: 0.5247 - val_acc: 0.7475
Epoch 15/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5225 - acc: 0.7475 - val_loss: 0.5307 - val_acc: 0.7437
Epoch 16/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5212 - acc: 0.7476 - val_loss: 0.5321 - val_acc: 0.7429
Epoch 17/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5208 - acc: 0.7485 - val_loss: 0.5258 - val_acc: 0.7450
Epoch 18/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5198 - acc: 0.7484 - val_loss: 0.5278 - val_acc: 0.7457
Epoch 19/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5184 - acc: 0.7496 - val_loss: 0.5233 - val_acc: 0.7486
Epoch 20/100
390870/390870 [==============================] - 51s 130us/step - loss: 0.5182 - acc: 0.7496 - val_loss: 0.5240 - val_acc: 0.7476
Epoch 21/100
390870/390870 [==============================] - 50s 127us/step - loss: 0.5179 - acc: 0.7498 - val_loss: 0.5236 - val_acc: 0.7484
Epoch 22/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5163 - acc: 0.7512 - val_loss: 0.5221 - val_acc: 0.7483
Epoch 23/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5160 - acc: 0.7514 - val_loss: 0.5288 - val_acc: 0.7450
Epoch 24/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5153 - acc: 0.7514 - val_loss: 0.5319 - val_acc: 0.7449
Epoch 25/100
390870/390870 [==============================] - 56s 144us/step - loss: 0.5151 - acc: 0.7520 - val_loss: 0.5248 - val_acc: 0.7468
Epoch 26/100
390870/390870 [==============================] - 122s 311us/step - loss: 0.5141 - acc: 0.7520 - val_loss: 0.5198 - val_acc: 0.7498
Epoch 27/100
390870/390870 [==============================] - 53s 136us/step - loss: 0.5146 - acc: 0.7514 - val_loss: 0.5223 - val_acc: 0.7492
Epoch 28/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5138 - acc: 0.7523 - val_loss: 0.5232 - val_acc: 0.7480
Epoch 29/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5123 - acc: 0.7533 - val_loss: 0.5196 - val_acc: 0.7512
Epoch 30/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5130 - acc: 0.7525 - val_loss: 0.5197 - val_acc: 0.7503
Epoch 31/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5119 - acc: 0.7541 - val_loss: 0.5242 - val_acc: 0.7493
Epoch 32/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5123 - acc: 0.7531 - val_loss: 0.5248 - val_acc: 0.7480
Epoch 33/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5113 - acc: 0.7539 - val_loss: 0.5227 - val_acc: 0.7486
Epoch 34/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5113 - acc: 0.7541 - val_loss: 0.5208 - val_acc: 0.7507
Epoch 35/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5114 - acc: 0.7531 - val_loss: 0.5202 - val_acc: 0.7488
Epoch 36/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5106 - acc: 0.7544 - val_loss: 0.5245 - val_acc: 0.7461
Epoch 37/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5109 - acc: 0.7537 - val_loss: 0.5253 - val_acc: 0.7459
Epoch 38/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5099 - acc: 0.7546 - val_loss: 0.5179 - val_acc: 0.7513
Epoch 39/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5088 - acc: 0.7551 - val_loss: 0.5195 - val_acc: 0.7495
Epoch 40/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5093 - acc: 0.7549 - val_loss: 0.5187 - val_acc: 0.7514
Epoch 41/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5084 - acc: 0.7555 - val_loss: 0.5247 - val_acc: 0.7485
Epoch 42/100
390870/390870 [==============================] - 51s 131us/step - loss: 0.5084 - acc: 0.7553 - val_loss: 0.5258 - val_acc: 0.7479
Epoch 43/100
390870/390870 [==============================] - 52s 134us/step - loss: 0.5077 - acc: 0.7554 - val_loss: 0.5195 - val_acc: 0.7511
Epoch 44/100
390870/390870 [==============================] - 56s 144us/step - loss: 0.5073 - acc: 0.7555 - val_loss: 0.5191 - val_acc: 0.7501
Epoch 45/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5077 - acc: 0.7554 - val_loss: 0.5202 - val_acc: 0.7508
Epoch 46/100
390870/390870 [==============================] - 52s 134us/step - loss: 0.5069 - acc: 0.7567 - val_loss: 0.5241 - val_acc: 0.7480
Epoch 47/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5073 - acc: 0.7561 - val_loss: 0.5229 - val_acc: 0.7477
Epoch 48/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5076 - acc: 0.7555 - val_loss: 0.5235 - val_acc: 0.7488
Epoch 49/100
390870/390870 [==============================] - 53s 134us/step - loss: 0.5053 - acc: 0.7571 - val_loss: 0.5196 - val_acc: 0.7500
Epoch 50/100
390870/390870 [==============================] - 202s 517us/step - loss: 0.5063 - acc: 0.7562 - val_loss: 0.5226 - val_acc: 0.7473
Epoch 51/100
390870/390870 [==============================] - 51s 129us/step - loss: 0.5055 - acc: 0.7571 - val_loss: 0.5298 - val_acc: 0.7420
Epoch 52/100
390870/390870 [==============================] - 51s 130us/step - loss: 0.5054 - acc: 0.7571 - val_loss: 0.5174 - val_acc: 0.7514
Epoch 53/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5054 - acc: 0.7568 - val_loss: 0.5236 - val_acc: 0.7496
Epoch 54/100
390870/390870 [==============================] - 50s 127us/step - loss: 0.5045 - acc: 0.7576 - val_loss: 0.5235 - val_acc: 0.7474
Epoch 55/100
390870/390870 [==============================] - 59s 150us/step - loss: 0.5046 - acc: 0.7576 - val_loss: 0.5263 - val_acc: 0.7475
Epoch 56/100
390870/390870 [==============================] - 61s 156us/step - loss: 0.5040 - acc: 0.7581 - val_loss: 0.5194 - val_acc: 0.7503
Epoch 57/100
390870/390870 [==============================] - 55s 142us/step - loss: 0.5051 - acc: 0.7571 - val_loss: 0.5175 - val_acc: 0.7527
Epoch 58/100
390870/390870 [==============================] - 58s 147us/step - loss: 0.5044 - acc: 0.7573 - val_loss: 0.5236 - val_acc: 0.7504
Epoch 59/100
390870/390870 [==============================] - 53s 136us/step - loss: 0.5040 - acc: 0.7578 - val_loss: 0.5192 - val_acc: 0.7520
Epoch 60/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5034 - acc: 0.7588 - val_loss: 0.5231 - val_acc: 0.7494
Epoch 61/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5031 - acc: 0.7582 - val_loss: 0.5236 - val_acc: 0.7493
Epoch 62/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5036 - acc: 0.7582 - val_loss: 0.5184 - val_acc: 0.7514
Epoch 63/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5038 - acc: 0.7582 - val_loss: 0.5226 - val_acc: 0.7502
Epoch 64/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5035 - acc: 0.7580 - val_loss: 0.5202 - val_acc: 0.7512
Epoch 65/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5029 - acc: 0.7586 - val_loss: 0.5237 - val_acc: 0.7473
Epoch 66/100
390870/390870 [==============================] - 52s 134us/step - loss: 0.5023 - acc: 0.7586 - val_loss: 0.5242 - val_acc: 0.7488
Epoch 67/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5031 - acc: 0.7582 - val_loss: 0.5237 - val_acc: 0.7483
Epoch 68/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5023 - acc: 0.7591 - val_loss: 0.5277 - val_acc: 0.7460
Epoch 69/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5024 - acc: 0.7587 - val_loss: 0.5266 - val_acc: 0.7505
Epoch 70/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5028 - acc: 0.7586 - val_loss: 0.5228 - val_acc: 0.7484
Epoch 71/100
390870/390870 [==============================] - 50s 127us/step - loss: 0.5015 - acc: 0.7593 - val_loss: 0.5199 - val_acc: 0.7515
Epoch 72/100
390870/390870 [==============================] - 50s 127us/step - loss: 0.5018 - acc: 0.7589 - val_loss: 0.5213 - val_acc: 0.7502
Epoch 73/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.5016 - acc: 0.7593 - val_loss: 0.5244 - val_acc: 0.7480
Epoch 74/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5013 - acc: 0.7595 - val_loss: 0.5251 - val_acc: 0.7483
Epoch 75/100
390870/390870 [==============================] - 54s 138us/step - loss: 0.5010 - acc: 0.7599 - val_loss: 0.5296 - val_acc: 0.7451
Epoch 76/100
390870/390870 [==============================] - 55s 140us/step - loss: 0.5010 - acc: 0.7598 - val_loss: 0.5251 - val_acc: 0.7493
Epoch 77/100
390870/390870 [==============================] - 64s 163us/step - loss: 0.5003 - acc: 0.7599 - val_loss: 0.5200 - val_acc: 0.7535
Epoch 78/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.5009 - acc: 0.7596 - val_loss: 0.5217 - val_acc: 0.7486
Epoch 79/100
390870/390870 [==============================] - 68s 173us/step - loss: 0.5000 - acc: 0.7596 - val_loss: 0.5240 - val_acc: 0.7490
Epoch 80/100
390870/390870 [==============================] - 76s 196us/step - loss: 0.5000 - acc: 0.7600 - val_loss: 0.5238 - val_acc: 0.7480
Epoch 81/100
390870/390870 [==============================] - 169s 433us/step - loss: 0.4997 - acc: 0.7599 - val_loss: 0.5271 - val_acc: 0.7491
Epoch 82/100
390870/390870 [==============================] - 59s 150us/step - loss: 0.4992 - acc: 0.7605 - val_loss: 0.5217 - val_acc: 0.7517
Epoch 83/100
390870/390870 [==============================] - 91s 234us/step - loss: 0.4992 - acc: 0.7603 - val_loss: 0.5226 - val_acc: 0.7511
Epoch 84/100
390870/390870 [==============================] - 69s 177us/step - loss: 0.5003 - acc: 0.7594 - val_loss: 0.5291 - val_acc: 0.7458
Epoch 85/100
390870/390870 [==============================] - 50s 128us/step - loss: 0.4991 - acc: 0.7601 - val_loss: 0.5210 - val_acc: 0.7500
Epoch 86/100
390870/390870 [==============================] - 51s 131us/step - loss: 0.4992 - acc: 0.7599 - val_loss: 0.5225 - val_acc: 0.7509
Epoch 87/100
390870/390870 [==============================] - 52s 133us/step - loss: 0.4989 - acc: 0.7604 - val_loss: 0.5234 - val_acc: 0.7493
Epoch 88/100
390870/390870 [==============================] - 53s 137us/step - loss: 0.4979 - acc: 0.7608 - val_loss: 0.5315 - val_acc: 0.7489
Epoch 89/100
390870/390870 [==============================] - 61s 156us/step - loss: 0.4985 - acc: 0.7607 - val_loss: 0.5237 - val_acc: 0.7487
Epoch 90/100
390870/390870 [==============================] - 74s 188us/step - loss: 0.4992 - acc: 0.7602 - val_loss: 0.5223 - val_acc: 0.7511
Epoch 91/100
390870/390870 [==============================] - 85s 216us/step - loss: 0.4984 - acc: 0.7611 - val_loss: 0.5245 - val_acc: 0.7499
Epoch 92/100
390870/390870 [==============================] - 67s 172us/step - loss: 0.4972 - acc: 0.7615 - val_loss: 0.5271 - val_acc: 0.7492
Epoch 93/100
390870/390870 [==============================] - 61s 155us/step - loss: 0.4976 - acc: 0.7611 - val_loss: 0.5268 - val_acc: 0.7457
Epoch 94/100
390870/390870 [==============================] - 60s 152us/step - loss: 0.4977 - acc: 0.7612 - val_loss: 0.5360 - val_acc: 0.7415
Epoch 95/100
390870/390870 [==============================] - 60s 153us/step - loss: 0.4979 - acc: 0.7608 - val_loss: 0.5229 - val_acc: 0.7495
Epoch 96/100
390870/390870 [==============================] - 54s 138us/step - loss: 0.4976 - acc: 0.7613 - val_loss: 0.5265 - val_acc: 0.7504
Epoch 97/100
390870/390870 [==============================] - 56s 142us/step - loss: 0.4971 - acc: 0.7614 - val_loss: 0.5264 - val_acc: 0.7495
Epoch 98/100
390870/390870 [==============================] - 58s 147us/step - loss: 0.4969 - acc: 0.7613 - val_loss: 0.5283 - val_acc: 0.7470
Epoch 99/100
390870/390870 [==============================] - 50s 127us/step - loss: 0.4975 - acc: 0.7614 - val_loss: 0.5328 - val_acc: 0.7460
Epoch 100/100
390870/390870 [==============================] - 50s 127us/step - loss: 0.4967 - acc: 0.7620 - val_loss: 0.5406 - val_acc: 0.7454

<keras.callbacks.History at 0x2859abf6860>

# Save trained model
classifier.save("../models/trained_deep_neural_network_all8.h5")

# evaluation of the model training
classifier = _load_model("../models/trained_deep_neural_network_all8.h5")  # DEBUG!!!
evaluation = classifier.evaluate(train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])
print(evaluation)

558386/558386 [==============================] - 54s 97us/step
[0.5078941215270898, 0.7580132739713417]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_2.pickle')
test_set.shape

(153993, 2774)

prediction = classifier.predict(test_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
binary_prediction = _binarize(prediction, threshold = 0.5)
binary_prediction

array([[0.],
       [0.],
       [1.],
       ...,
       [1.],
       [1.],
       [1.]], dtype=float32)

set(binary_prediction[:,0])

{0.0, 1.0}

print("Report for Deep Learning approach:")
print(_classification_report(test_set['likes'], binary_prediction))

print("Accuracy for Deep Learning approach:", _accuracy_score(test_set['likes'], binary_prediction) * 100)

Report for Deep Learning approach:
              precision    recall  f1-score   support

           0       0.67      0.44      0.53     50930
           1       0.76      0.89      0.82    103063

    accuracy                           0.74    153993
   macro avg       0.72      0.67      0.68    153993
weighted avg       0.73      0.74      0.73    153993

Accuracy for Deep Learning approach: 74.25792081458248

matrix = _confusion_matrix(test_set['likes'], binary_prediction)
print(matrix)

[[22257 28673]
 [10968 92095]]

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], binary_prediction)

_plt.plot(fpr, tpr)
_plt.xlim([0.0, 1.0])
_plt.ylim([0.0, 1.0])

_plt.title("Neural Network ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_del_all()

NN 8¶

alpha = 8

# hidden layers = 5

batch size = 100

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_2.pickle')
train_set.shape

(558386, 2774)

# Number of neurons for hiddens layer
number_samples, number_features = train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']).shape

number_input_neurons = number_features
number_output_neurons = 1
number_train_semples = number_samples
alpha = 8

number_hidden_neurons = round(number_train_semples / (alpha * (number_input_neurons + number_output_neurons)))
print(number_hidden_neurons)

25

classifier = _Sequential()

# First Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal', input_dim = number_features))

# Second Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Third Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fourth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fifth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Output Layer
classifier.add(_Dense(1, activation = 'sigmoid', kernel_initializer = 'random_normal'))

# Compiling the neural network
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

# Fitting the data to the training dataset
classifier.fit(train_set.drop(
    columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']),
    train_set['likes'], validation_split = 0.3, batch_size = 100, epochs = 100)

Train on 390870 samples, validate on 167516 samples
Epoch 1/100
390870/390870 [==============================] - 97s 249us/step - loss: 0.5884 - acc: 0.6991 - val_loss: 0.5596 - val_acc: 0.7209
Epoch 2/100
390870/390870 [==============================] - 132s 339us/step - loss: 0.5530 - acc: 0.7257 - val_loss: 0.5444 - val_acc: 0.7335
Epoch 3/100
390870/390870 [==============================] - 92s 235us/step - loss: 0.5437 - acc: 0.7323 - val_loss: 0.5402 - val_acc: 0.7374
Epoch 4/100
390870/390870 [==============================] - 80s 205us/step - loss: 0.5398 - acc: 0.7353 - val_loss: 0.5386 - val_acc: 0.7420
Epoch 5/100
390870/390870 [==============================] - 76s 194us/step - loss: 0.5376 - acc: 0.7369 - val_loss: 0.5458 - val_acc: 0.7319
Epoch 6/100
390870/390870 [==============================] - 78s 198us/step - loss: 0.5349 - acc: 0.7388 - val_loss: 0.5438 - val_acc: 0.7389
Epoch 7/100
390870/390870 [==============================] - 76s 195us/step - loss: 0.5334 - acc: 0.7397 - val_loss: 0.5317 - val_acc: 0.7416
Epoch 8/100
390870/390870 [==============================] - 74s 190us/step - loss: 0.5311 - acc: 0.7415 - val_loss: 0.5345 - val_acc: 0.7405
Epoch 9/100
390870/390870 [==============================] - 76s 195us/step - loss: 0.5299 - acc: 0.7424 - val_loss: 0.5317 - val_acc: 0.7461
Epoch 10/100
390870/390870 [==============================] - 81s 206us/step - loss: 0.5287 - acc: 0.7425 - val_loss: 0.5283 - val_acc: 0.7460
Epoch 11/100
390870/390870 [==============================] - 95s 244us/step - loss: 0.5276 - acc: 0.7439 - val_loss: 0.5367 - val_acc: 0.7427
Epoch 12/100
390870/390870 [==============================] - 78s 199us/step - loss: 0.5263 - acc: 0.7445 - val_loss: 0.5266 - val_acc: 0.7483
Epoch 13/100
390870/390870 [==============================] - 75s 193us/step - loss: 0.5253 - acc: 0.7453 - val_loss: 0.5278 - val_acc: 0.7463
Epoch 14/100
390870/390870 [==============================] - 76s 195us/step - loss: 0.5242 - acc: 0.7459 - val_loss: 0.5301 - val_acc: 0.7438
Epoch 15/100
390870/390870 [==============================] - 80s 206us/step - loss: 0.5239 - acc: 0.7464 - val_loss: 0.5226 - val_acc: 0.7489
Epoch 16/100
390870/390870 [==============================] - 74s 188us/step - loss: 0.5224 - acc: 0.7475 - val_loss: 0.5273 - val_acc: 0.7468
Epoch 17/100
390870/390870 [==============================] - 74s 190us/step - loss: 0.5222 - acc: 0.7471 - val_loss: 0.5276 - val_acc: 0.7452
Epoch 18/100
390870/390870 [==============================] - 72s 184us/step - loss: 0.5216 - acc: 0.7483 - val_loss: 0.5333 - val_acc: 0.7430
Epoch 19/100
390870/390870 [==============================] - 72s 185us/step - loss: 0.5208 - acc: 0.7486 - val_loss: 0.5300 - val_acc: 0.7431
Epoch 20/100
390870/390870 [==============================] - 72s 184us/step - loss: 0.5201 - acc: 0.7489 - val_loss: 0.5244 - val_acc: 0.7486
Epoch 21/100
390870/390870 [==============================] - 72s 185us/step - loss: 0.5198 - acc: 0.7493 - val_loss: 0.5246 - val_acc: 0.7478
Epoch 22/100
390870/390870 [==============================] - 73s 186us/step - loss: 0.5195 - acc: 0.7493 - val_loss: 0.5225 - val_acc: 0.7476
Epoch 23/100
390870/390870 [==============================] - 72s 185us/step - loss: 0.5189 - acc: 0.7499 - val_loss: 0.5311 - val_acc: 0.7405
Epoch 24/100
390870/390870 [==============================] - 74s 189us/step - loss: 0.5188 - acc: 0.7495 - val_loss: 0.5260 - val_acc: 0.7463
Epoch 25/100
390870/390870 [==============================] - 74s 190us/step - loss: 0.5180 - acc: 0.7506 - val_loss: 0.5273 - val_acc: 0.7459
Epoch 26/100
390870/390870 [==============================] - 73s 187us/step - loss: 0.5176 - acc: 0.7511 - val_loss: 0.5262 - val_acc: 0.7446
Epoch 27/100
390870/390870 [==============================] - 73s 186us/step - loss: 0.5171 - acc: 0.7505 - val_loss: 0.5367 - val_acc: 0.7423
Epoch 28/100
390870/390870 [==============================] - 92s 236us/step - loss: 0.5163 - acc: 0.7516 - val_loss: 0.5235 - val_acc: 0.7469
Epoch 29/100
390870/390870 [==============================] - 78s 198us/step - loss: 0.5163 - acc: 0.7515 - val_loss: 0.5213 - val_acc: 0.7492
Epoch 30/100
390870/390870 [==============================] - 76s 195us/step - loss: 0.5162 - acc: 0.7514 - val_loss: 0.5226 - val_acc: 0.7480
Epoch 31/100
390870/390870 [==============================] - 72s 185us/step - loss: 0.5155 - acc: 0.7518 - val_loss: 0.5293 - val_acc: 0.7445
Epoch 32/100
390870/390870 [==============================] - 73s 187us/step - loss: 0.5146 - acc: 0.7524 - val_loss: 0.5318 - val_acc: 0.7457
Epoch 33/100
390870/390870 [==============================] - 73s 186us/step - loss: 0.5144 - acc: 0.7521 - val_loss: 0.5209 - val_acc: 0.7499
Epoch 34/100
390870/390870 [==============================] - 72s 184us/step - loss: 0.5143 - acc: 0.7529 - val_loss: 0.5211 - val_acc: 0.7481
Epoch 35/100
390870/390870 [==============================] - 73s 185us/step - loss: 0.5134 - acc: 0.7530 - val_loss: 0.5218 - val_acc: 0.7470
Epoch 36/100
390870/390870 [==============================] - 75s 193us/step - loss: 0.5132 - acc: 0.7532 - val_loss: 0.5192 - val_acc: 0.7500
Epoch 37/100
390870/390870 [==============================] - 77s 196us/step - loss: 0.5128 - acc: 0.7533 - val_loss: 0.5198 - val_acc: 0.7502
Epoch 38/100
390870/390870 [==============================] - 77s 197us/step - loss: 0.5133 - acc: 0.7531 - val_loss: 0.5205 - val_acc: 0.7489
Epoch 39/100
390870/390870 [==============================] - 77s 197us/step - loss: 0.5127 - acc: 0.7534 - val_loss: 0.5202 - val_acc: 0.7491
Epoch 40/100
390870/390870 [==============================] - 82s 210us/step - loss: 0.5122 - acc: 0.7536 - val_loss: 0.5198 - val_acc: 0.7494
Epoch 41/100
390870/390870 [==============================] - 77s 197us/step - loss: 0.5120 - acc: 0.7536 - val_loss: 0.5220 - val_acc: 0.7478
Epoch 42/100
390870/390870 [==============================] - 76s 195us/step - loss: 0.5120 - acc: 0.7537 - val_loss: 0.5203 - val_acc: 0.7497
Epoch 43/100
390870/390870 [==============================] - 77s 196us/step - loss: 0.5117 - acc: 0.7541 - val_loss: 0.5228 - val_acc: 0.7490
Epoch 44/100
390870/390870 [==============================] - 74s 188us/step - loss: 0.5109 - acc: 0.7545 - val_loss: 0.5198 - val_acc: 0.7495
Epoch 45/100
390870/390870 [==============================] - 73s 186us/step - loss: 0.5139 - acc: 0.7526 - val_loss: 0.5190 - val_acc: 0.7505
Epoch 46/100
390870/390870 [==============================] - 73s 188us/step - loss: 0.5104 - acc: 0.7548 - val_loss: 0.5273 - val_acc: 0.7465
Epoch 47/100
390870/390870 [==============================] - 73s 187us/step - loss: 0.5104 - acc: 0.7545 - val_loss: 0.5218 - val_acc: 0.7480
Epoch 48/100
390870/390870 [==============================] - 79s 202us/step - loss: 0.5100 - acc: 0.7545 - val_loss: 0.5193 - val_acc: 0.7506
Epoch 49/100
390870/390870 [==============================] - 78s 200us/step - loss: 0.5098 - acc: 0.7549 - val_loss: 0.5275 - val_acc: 0.7479
Epoch 50/100
390870/390870 [==============================] - 90s 230us/step - loss: 0.5096 - acc: 0.7550 - val_loss: 0.5255 - val_acc: 0.7474
Epoch 51/100
390870/390870 [==============================] - 100s 255us/step - loss: 0.5093 - acc: 0.7553 - val_loss: 0.5207 - val_acc: 0.7522
Epoch 52/100
390870/390870 [==============================] - 79s 201us/step - loss: 0.5093 - acc: 0.7549 - val_loss: 0.5196 - val_acc: 0.7514
Epoch 53/100
390870/390870 [==============================] - 90s 230us/step - loss: 0.5086 - acc: 0.7554 - val_loss: 0.5198 - val_acc: 0.7516
Epoch 54/100
390870/390870 [==============================] - 91s 232us/step - loss: 0.5092 - acc: 0.7558 - val_loss: 0.5237 - val_acc: 0.7477
Epoch 55/100
390870/390870 [==============================] - 103s 265us/step - loss: 0.5085 - acc: 0.7556 - val_loss: 0.5259 - val_acc: 0.7481
Epoch 56/100
390870/390870 [==============================] - 112s 287us/step - loss: 0.5084 - acc: 0.7555 - val_loss: 0.5217 - val_acc: 0.7498
Epoch 57/100
390870/390870 [==============================] - 94s 241us/step - loss: 0.5083 - acc: 0.7555 - val_loss: 0.5194 - val_acc: 0.7516
Epoch 58/100
390870/390870 [==============================] - 89s 228us/step - loss: 0.5079 - acc: 0.7563 - val_loss: 0.5220 - val_acc: 0.7506
Epoch 59/100
390870/390870 [==============================] - 82s 210us/step - loss: 0.5080 - acc: 0.7557 - val_loss: 0.5229 - val_acc: 0.7503
Epoch 60/100
390870/390870 [==============================] - 77s 196us/step - loss: 0.5073 - acc: 0.7559 - val_loss: 0.5187 - val_acc: 0.7518
Epoch 61/100
390870/390870 [==============================] - 77s 197us/step - loss: 0.5070 - acc: 0.7566 - val_loss: 0.5272 - val_acc: 0.7465
Epoch 62/100
390870/390870 [==============================] - 85s 217us/step - loss: 0.5070 - acc: 0.7570 - val_loss: 0.5198 - val_acc: 0.7506
Epoch 63/100
390870/390870 [==============================] - 101s 258us/step - loss: 0.5067 - acc: 0.7566 - val_loss: 0.5224 - val_acc: 0.7503
Epoch 64/100
390870/390870 [==============================] - 113s 289us/step - loss: 0.5063 - acc: 0.7567 - val_loss: 0.5261 - val_acc: 0.7476
Epoch 65/100
390870/390870 [==============================] - 117s 299us/step - loss: 0.5071 - acc: 0.7564 - val_loss: 0.5200 - val_acc: 0.7499
Epoch 66/100
390870/390870 [==============================] - 111s 285us/step - loss: 0.5063 - acc: 0.7565 - val_loss: 0.5235 - val_acc: 0.7477
Epoch 67/100
390870/390870 [==============================] - 88s 226us/step - loss: 0.5065 - acc: 0.7561 - val_loss: 0.5248 - val_acc: 0.7497
Epoch 68/100
390870/390870 [==============================] - 89s 228us/step - loss: 0.5064 - acc: 0.7562 - val_loss: 0.5201 - val_acc: 0.7511
Epoch 69/100
390870/390870 [==============================] - 82s 209us/step - loss: 0.5064 - acc: 0.7568 - val_loss: 0.5265 - val_acc: 0.7486
Epoch 70/100
390870/390870 [==============================] - 82s 209us/step - loss: 0.5062 - acc: 0.7570 - val_loss: 0.5222 - val_acc: 0.7506
Epoch 71/100
390870/390870 [==============================] - 80s 206us/step - loss: 0.5058 - acc: 0.7564 - val_loss: 0.5197 - val_acc: 0.7507
Epoch 72/100
390870/390870 [==============================] - 84s 214us/step - loss: 0.5055 - acc: 0.7573 - val_loss: 0.5231 - val_acc: 0.7489
Epoch 73/100
390870/390870 [==============================] - 95s 244us/step - loss: 0.5052 - acc: 0.7573 - val_loss: 0.5216 - val_acc: 0.7492
Epoch 74/100
390870/390870 [==============================] - 100s 257us/step - loss: 0.5052 - acc: 0.7576 - val_loss: 0.5250 - val_acc: 0.7475
Epoch 75/100
390870/390870 [==============================] - 99s 253us/step - loss: 0.5055 - acc: 0.7571 - val_loss: 0.5184 - val_acc: 0.7519
Epoch 76/100
390870/390870 [==============================] - 84s 215us/step - loss: 0.5047 - acc: 0.7576 - val_loss: 0.5311 - val_acc: 0.7466
Epoch 77/100
390870/390870 [==============================] - 87s 223us/step - loss: 0.5048 - acc: 0.7577 - val_loss: 0.5252 - val_acc: 0.7476
Epoch 78/100
390870/390870 [==============================] - 84s 215us/step - loss: 0.5054 - acc: 0.7566 - val_loss: 0.5271 - val_acc: 0.7445
Epoch 79/100
390870/390870 [==============================] - 101s 259us/step - loss: 0.5045 - acc: 0.7573 - val_loss: 0.5220 - val_acc: 0.7510
Epoch 80/100
390870/390870 [==============================] - 107s 274us/step - loss: 0.5043 - acc: 0.7575 - val_loss: 0.5237 - val_acc: 0.7477
Epoch 81/100
390870/390870 [==============================] - 98s 250us/step - loss: 0.5042 - acc: 0.7577 - val_loss: 0.5262 - val_acc: 0.7489
Epoch 82/100
390870/390870 [==============================] - 101s 259us/step - loss: 0.5048 - acc: 0.7574 - val_loss: 0.5237 - val_acc: 0.7488
Epoch 83/100
390870/390870 [==============================] - 100s 256us/step - loss: 0.5037 - acc: 0.7581 - val_loss: 0.5210 - val_acc: 0.7500
Epoch 84/100
390870/390870 [==============================] - 82s 209us/step - loss: 0.5040 - acc: 0.7577 - val_loss: 0.5231 - val_acc: 0.7492
Epoch 85/100
390870/390870 [==============================] - 90s 230us/step - loss: 0.5039 - acc: 0.7577 - val_loss: 0.5263 - val_acc: 0.7485
Epoch 86/100
390870/390870 [==============================] - 81s 208us/step - loss: 0.5040 - acc: 0.7578 - val_loss: 0.5234 - val_acc: 0.7503
Epoch 87/100
390870/390870 [==============================] - 82s 211us/step - loss: 0.5032 - acc: 0.7582 - val_loss: 0.5262 - val_acc: 0.7494
Epoch 88/100
390870/390870 [==============================] - 84s 214us/step - loss: 0.5031 - acc: 0.7586 - val_loss: 0.5333 - val_acc: 0.7450
Epoch 89/100
390870/390870 [==============================] - 91s 233us/step - loss: 0.5028 - acc: 0.7583 - val_loss: 0.5206 - val_acc: 0.7496
Epoch 90/100
390870/390870 [==============================] - 87s 223us/step - loss: 0.5024 - acc: 0.7590 - val_loss: 0.5294 - val_acc: 0.7455
Epoch 91/100
390870/390870 [==============================] - 87s 223us/step - loss: 0.5033 - acc: 0.7585 - val_loss: 0.5391 - val_acc: 0.7455
Epoch 92/100
390870/390870 [==============================] - 87s 223us/step - loss: 0.5028 - acc: 0.7586 - val_loss: 0.5250 - val_acc: 0.7494
Epoch 93/100
390870/390870 [==============================] - 92s 234us/step - loss: 0.5032 - acc: 0.7589 - val_loss: 0.5213 - val_acc: 0.7506
Epoch 94/100
390870/390870 [==============================] - 87s 222us/step - loss: 0.5024 - acc: 0.7588 - val_loss: 0.5243 - val_acc: 0.7493
Epoch 95/100
390870/390870 [==============================] - 84s 216us/step - loss: 0.5023 - acc: 0.7582 - val_loss: 0.5226 - val_acc: 0.7497
Epoch 96/100
390870/390870 [==============================] - 86s 220us/step - loss: 0.5026 - acc: 0.7583 - val_loss: 0.5276 - val_acc: 0.7455
Epoch 97/100
390870/390870 [==============================] - 82s 209us/step - loss: 0.5020 - acc: 0.7587 - val_loss: 0.5239 - val_acc: 0.7504
Epoch 98/100
390870/390870 [==============================] - 78s 200us/step - loss: 0.5022 - acc: 0.7591 - val_loss: 0.5307 - val_acc: 0.7451
Epoch 99/100
390870/390870 [==============================] - 74s 190us/step - loss: 0.5018 - acc: 0.7589 - val_loss: 0.5308 - val_acc: 0.7447
Epoch 100/100
390870/390870 [==============================] - 87s 222us/step - loss: 0.5023 - acc: 0.7592 - val_loss: 0.5284 - val_acc: 0.7470

<keras.callbacks.History at 0x19a25de07b8>

# Save trained model
classifier.save("../models/trained_deep_neural_network_all9.h5")

# evaluation of the model training
classifier = _load_model("../models/trained_deep_neural_network_all8.h5")  # DEBUG!!!
evaluation = classifier.evaluate(train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])
print(evaluation)

558386/558386 [==============================] - 58s 104us/step
[0.5078941215270898, 0.7580132739713417]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_2.pickle')
test_set.shape

(153993, 2774)

prediction = classifier.predict(test_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
binary_prediction = _binarize(prediction, threshold = 0.5)
binary_prediction

array([[0.],
       [0.],
       [1.],
       ...,
       [1.],
       [1.],
       [1.]], dtype=float32)

set(binary_prediction[:,0])

{0.0, 1.0}

print("Report for Deep Learning approach:")
print(_classification_report(test_set['likes'], binary_prediction))

print("Accuracy for Deep Learning approach:", _accuracy_score(test_set['likes'], binary_prediction) * 100)

Report for Deep Learning approach:
              precision    recall  f1-score   support

           0       0.67      0.44      0.53     50930
           1       0.76      0.89      0.82    103063

    accuracy                           0.74    153993
   macro avg       0.72      0.67      0.68    153993
weighted avg       0.73      0.74      0.73    153993

Accuracy for Deep Learning approach: 74.25792081458248

matrix = _confusion_matrix(test_set['likes'], binary_prediction)
print(matrix)

[[22257 28673]
 [10968 92095]]

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], binary_prediction)

_plt.plot(fpr, tpr)
_plt.xlim([0.0, 1.0])
_plt.ylim([0.0, 1.0])

_plt.title("Neural Network ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_del_all()

7.4 SVM without fake reviews¶

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')
train_set.shape

(558386, 787)

train_set = train_set[train_set['bin_truth_score']!=-1]
train_set.shape

(438571, 787)

best_model = _jl.load("../models/best_SVM.joblib")
best_model.set_params(verbose=10)
best_model.get_params()

{'C': 0.001,
 'class_weight': None,
 'dual': True,
 'fit_intercept': True,
 'intercept_scaling': 1,
 'loss': 'squared_hinge',
 'max_iter': 50000,
 'multi_class': 'ovr',
 'penalty': 'l2',
 'random_state': 0,
 'tol': 0.0001,
 'verbose': 10}

best_model.fit(train_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])

[LibLinear]

C:\Users\super\Anaconda3\envs\tf-gpu\lib\site-packages\sklearn\svm\base.py:929: ConvergenceWarning: Liblinear failed to converge, increase the number of iterations.
  "the number of iterations.", ConvergenceWarning)

LinearSVC(C=0.001, class_weight=None, dual=True, fit_intercept=True,
          intercept_scaling=1, loss='squared_hinge', max_iter=50000,
          multi_class='ovr', penalty='l2', random_state=0, tol=0.0001,
          verbose=10)

_jl.dump(best_model, "../models/best_SVM_fake.joblib")

['../models/best_SVM_fake.joblib']

print("coef:", best_model.coef_)
print("intercept:", best_model.intercept_)

coef: [[-1.38515487e-01 -1.96857288e-01  3.32402060e-01 -2.91255272e-01
  -1.69144522e-01  4.39209749e-02  2.42195943e-02  3.64156997e-02
  -2.46790916e-02 -1.02787607e-02 -4.86873857e-02 -4.97584956e-05
  -3.22987927e-02  6.88759948e-03 -1.03371270e-05  6.50742395e-01
  -1.29562758e-04  1.03416295e-03 -2.48775106e-04 -6.59642374e-04
  -1.02036028e-04 -9.66791010e-04  1.14359591e-03 -1.99843890e-04
  -6.27763021e-05  3.69227212e-04  2.38756561e-06 -1.18488072e-05
   3.94093254e-05 -2.29569450e-04 -8.67235837e-02  8.89035480e-06
  -7.57976137e-04  3.93742819e-04 -4.55403368e-02 -3.54663199e-04
  -3.18629560e-02  5.23979128e-02  3.66236285e-04  1.77877828e-02
  -8.52227099e-04 -2.36155968e-04  9.56668382e-05  1.75930155e-04
   6.77026802e-05 -5.13757145e-03  1.89471500e-02 -8.09871718e-04
   3.60054753e-03  6.19090568e-03  3.79168413e-02 -8.57468383e-04
   5.00425524e-02  5.05812751e-03  2.68403245e-02  4.03692064e-02
   2.11274126e-02  2.88136834e-02  1.73991590e-02 -5.80037087e-03
  -6.35825432e-03 -1.38553686e-03  3.66697695e-03  4.09770364e-03
   5.23367932e-03  1.44654196e-02 -2.06413754e-02 -1.05040811e-02
   1.54167703e-05 -3.50850188e-03 -5.80631608e-03  2.45154713e-02
  -1.89971779e-02  2.69824728e-02  2.13443201e-02 -4.47468512e-03
   9.90304353e-03  6.76647465e-03  3.02454848e-03  4.90134130e-02
  -1.07009586e-01 -7.27267822e-02 -1.11518903e-01 -5.34070255e-02
  -1.12431655e-01 -1.25416591e-01 -1.07294096e-01 -9.18106856e-02
  -9.21504902e-02 -1.02044568e-01 -9.66343841e-02 -9.25763200e-02
  -7.10907125e-02 -7.08271627e-02 -8.05048905e-02 -6.88325061e-02
  -7.62345358e-02 -7.94675628e-02 -7.16101280e-02 -6.39430451e-02
   3.83200703e-03  1.24178084e-02  9.68276063e-03  2.88133507e-02
   2.43884053e-02 -1.92121205e-02  1.52672097e-02  2.76503358e-02
  -6.89404188e-02  3.38157093e-02  3.36085218e-02 -5.62950213e-02
  -1.08435124e-02  2.23313546e-02  2.16739386e-02  6.37260612e-02
  -6.09296555e-03 -7.98115368e-02 -1.11556390e-02 -2.02394940e-02
   1.55625280e-03 -3.04833429e-03  3.51311190e-03  1.77761517e-02
   3.52897528e-03  1.50427985e-02  3.06900169e-01 -1.69078975e-02
   2.04658792e-02 -5.41781120e-03  8.74748899e-03  6.06713893e-03
   2.54657184e-02 -1.47421821e-02  2.64501303e-02 -1.68806851e-01
   2.52562977e-02  5.16353327e-02 -4.30719531e-03 -1.02730195e-01
   8.36980545e-03 -9.62089988e-02  1.72674874e-02 -1.51363526e-02
  -3.72019637e-02  1.67622388e-02  4.69927485e-03  1.19233652e-02
   1.03741758e-03  1.64732370e-02 -5.66227503e-02  1.45361303e-02
   7.48036703e-03  1.16303879e-02  2.17216512e-02 -9.05917181e-02
   1.36110522e-02 -8.78600518e-02 -2.75945059e-02 -5.04582035e-03
  -1.31594374e-02 -1.10745049e-02  5.87097288e-02  1.38382937e-02
   1.89250223e-02  1.63221636e-02 -8.19418169e-02  3.01171136e-02
  -3.14307123e-02 -5.22839544e-02  1.80781129e-02 -3.60306879e-02
   4.25120155e-02 -1.89666077e-02  4.26568614e-02  8.16088208e-02
   2.23159007e-02  1.49230095e-02  9.70061169e-03 -1.49506638e-02
   5.83322804e-03 -1.06623735e-02  1.22299549e-02  1.26321864e-01
  -1.95112136e-02 -2.17022819e-02 -1.08929688e-01  4.78160264e-03
  -1.76506556e-02  1.32192451e-02  2.26198213e-02  5.68965596e-02
  -2.50675783e-02  1.41512352e-02  1.12231983e-01 -2.30265518e-02
  -3.15793429e-02  1.51440539e-02 -2.18803563e-02 -1.51582937e-02
   3.79622761e-03  5.07480631e-02 -5.88344749e-03 -2.12778463e-02
  -7.30262693e-02  1.12750777e-02 -1.87664119e-02 -1.95641783e-02
   7.13358408e-02 -1.46709873e-02  1.83772840e-02 -4.37025997e-03
  -6.51153200e-02  2.46084236e-02  1.09030656e-04 -5.37629407e-02
   1.50915967e-02 -2.81888030e-03 -1.31207292e-02 -8.25417527e-03
  -3.86140908e-03 -2.36720403e-02  5.54096386e-02 -1.65074819e-02
  -1.05828789e-02  1.86338447e-02 -9.78510134e-02 -1.11608225e-02
  -1.10487593e-03  2.08611273e-02 -1.69727108e-03 -8.69724266e-02
   8.67924904e-03  1.64268003e-02 -4.08082984e-02 -2.02054386e-02
   3.43418891e-03 -2.89795457e-02  5.75343825e-02 -7.93273187e-03
   3.64269591e-02  1.05707660e-02 -1.98823667e-02  2.72527226e-02
   2.68108272e-02 -1.65154569e-02 -1.09612647e-02  3.97277385e-02
   1.45050380e-02 -5.12933921e-02 -4.22742148e-02 -3.46014737e-02
  -2.95818189e-02 -8.18524611e-02 -1.10265342e-02 -7.17019280e-03
   3.73694134e-04  8.19734824e-04  1.65446464e-02  1.25424032e-02
  -3.11586076e-02 -1.79967174e-02  6.84022293e-02 -1.35733274e-03
   2.44723270e-02 -3.34684862e-02 -3.22164793e-02  3.96808054e-03
  -1.86445809e-02 -8.67969553e-03  5.73984014e-03  9.37409516e-03
  -2.42438114e-02  1.89837635e-02 -7.64081435e-03 -4.67132415e-02
  -4.18300627e-03  2.05551077e-02 -4.88350793e-03 -1.86020479e-02
  -1.80049826e-02 -1.66074364e-02 -3.35520438e-03 -4.55070752e-02
   8.87395027e-03  1.28799348e-02  1.24229580e-02  1.12108721e-02
   1.45350581e-02 -2.29478080e-03 -5.51076001e-03 -2.28911425e-02
   7.48309204e-03 -6.79165167e-03  5.02895413e-03 -2.44546254e-03
  -4.87119098e-02  2.36266258e-03 -1.11751431e-03  4.33818702e-03
  -3.49420391e-03 -5.05476299e-02 -1.40039999e-03  3.27457911e-02
  -3.38286113e-03  9.16096697e-03 -4.68945199e-02  1.16020096e-02
  -6.04269370e-03  5.29924476e-03  1.32220965e-02  7.50497264e-03
  -7.88100646e-03 -1.75782857e-02 -6.37016406e-05 -1.00199407e-03
  -8.13866917e-04  1.88164705e-02 -1.28928581e-02 -5.23228929e-02
   1.84327185e-03  2.48377979e-02  4.08759215e-03  1.58755783e-02
  -3.76921166e-03  9.06916880e-03 -7.59932359e-03 -5.00339138e-02
   7.70580116e-03 -1.14642255e-02  3.66738698e-02 -3.86449439e-02
  -2.54845909e-02  1.63077941e-03  1.88310809e-02  1.74969020e-02
  -2.40183982e-02  8.83242193e-04 -8.71008343e-02 -7.57132392e-02
   2.26391977e-03  8.08202923e-03 -1.26930871e-02  3.25206351e-02
  -1.06429423e-02 -4.72443757e-03  4.36542243e-03 -3.30372496e-02
   5.77410389e-03 -2.22521881e-02  2.69829707e-02  1.39020942e-02
   3.90491981e-02 -8.28080995e-03  1.41291808e-02  4.47068889e-02
   4.21503603e-03  1.62920223e-02 -1.98530058e-02  2.11423809e-02
   5.04514238e-03 -8.08404004e-03 -2.71891009e-03  1.37625053e-02
   5.24562637e-02 -9.63055223e-04  1.17774946e-02  2.26308559e-03
   1.55985850e-02 -1.54669393e-02  2.70475234e-02  2.03787907e-02
  -2.21858211e-02  2.34364788e-04  5.68517038e-03 -8.54713483e-04
   1.54705314e-03 -2.35092407e-02  1.95500928e-02  9.16686779e-03
   8.20092446e-03  1.85429685e-02 -1.19026990e-02 -4.15252089e-02
  -9.96497140e-03  2.77428126e-02  2.31637304e-02 -8.63998865e-03
  -5.58969579e-03 -3.65804708e-02  1.02017273e-02  1.46447403e-02
  -5.32412891e-03  1.28184056e-02 -1.08280514e-03  4.64343466e-03
   2.68664654e-02  1.43696379e-02  4.50606194e-03  1.24790375e-02
   2.74475140e-03  3.26106722e-02 -6.94968408e-03  5.07915763e-03
  -5.32758708e-02 -2.07407535e-03  3.95504430e-03 -1.52842777e-02
  -1.47755630e-02  7.30146062e-03  8.55936075e-03 -1.25183829e-02
  -9.87121313e-03  1.83879276e-02 -1.34606118e-02 -2.59937612e-02
  -3.91079757e-02 -6.51453628e-03  5.07647070e-03 -1.35042537e-03
   4.98107954e-03 -1.77086859e-02 -3.51357363e-02  8.45129969e-03
  -3.29347257e-02  2.17426186e-02 -2.59214434e-02 -2.53633189e-02
   2.70112333e-02 -2.29994019e-02  2.15643648e-02  3.30667821e-02
  -1.21527344e-02 -8.04762544e-03  2.27048019e-03 -3.57113700e-02
  -1.95711470e-02  1.45399921e-02 -3.39962061e-02  5.11852896e-03
   7.12968111e-02 -9.61767322e-03  9.31277872e-03 -2.38795810e-02
  -1.13772699e-02 -1.01792396e-02  2.39231488e-03  6.68334194e-03
   1.55671785e-02  2.05874551e-02 -4.80607788e-03 -1.86829790e-02
  -1.45709490e-02 -1.06623722e-02  2.35326667e-02 -4.70131386e-03
  -4.21315624e-03 -2.71234831e-03  1.93383614e-03 -1.89008818e-02
   4.86325848e-03  5.09044025e-02  7.89412493e-03 -1.85020881e-02
  -1.65166310e-02  2.59124645e-02 -3.96222717e-03 -1.82581455e-03
  -2.84639480e-03 -6.21755254e-04  5.29924476e-03 -1.50841079e-02
  -4.39750044e-03 -4.42544188e-03  1.09647180e-02 -9.85470723e-03
   1.74375033e-02  4.02869950e-03 -9.80037647e-03  4.82269799e-03
  -5.86701112e-02 -9.05448823e-03  3.29424010e-02 -3.10830934e-02
  -4.07787452e-02  1.32919262e-02  6.53352095e-02 -5.48590234e-03
   1.02267075e-02  6.69731008e-03  2.74703725e-02  1.88225893e-02
   6.64172161e-02  9.56234199e-03  5.69613301e-03  1.00842961e-02
   2.02640525e-03  4.17134666e-03  8.80238582e-04 -9.58839652e-03
  -1.12144824e-02 -1.68011935e-02  1.76108325e-02 -2.49569806e-02
   2.10639397e-02  1.62382271e-02 -9.51530479e-03  3.35571782e-02
   8.82414610e-04 -6.01971158e-03  1.40010253e-02 -2.23258493e-03
   1.09818752e-02 -1.32184042e-02 -1.16853055e-02  7.75300062e-03
  -2.26713625e-03  2.36595861e-02 -2.33222683e-02 -5.46096270e-02
   4.16589791e-02 -1.03794442e-03 -3.14466420e-02  2.01603301e-02
  -2.96612174e-02  5.00838207e-03 -2.72989050e-02  2.37512453e-02
  -1.19723609e-02 -1.16283640e-03  1.77372095e-02 -1.47336557e-02
   1.31102188e-03  8.51543272e-03 -6.68587037e-03  1.99759053e-02
  -3.03666952e-02 -1.59449661e-02 -7.98116436e-03  6.09597181e-03
   3.07038472e-02  1.60983762e-03 -2.55458962e-02  2.57303489e-02
   3.98623009e-03 -2.49897178e-03 -4.61094740e-02  3.68646964e-02
   6.88660748e-03  1.87634909e-02  7.88533428e-03 -3.23301450e-03
  -5.52813662e-03 -1.78606036e-02 -1.79463803e-02 -3.57198186e-02
   2.53094829e-02  3.93922796e-03  1.61779464e-02  1.90456601e-02
   2.49921754e-02  1.21722342e-02  1.34071085e-03  1.45581362e-02
  -2.14437115e-03  2.24086577e-02 -4.53714331e-02 -2.74531767e-03
  -3.40563217e-03 -1.26440357e-02 -3.12196341e-02  1.86247094e-02
  -3.28087898e-03  1.82956600e-03 -3.64854486e-02  1.26395711e-02
   4.91543431e-03  1.50202129e-03 -4.86245017e-03  1.39062633e-03
   1.69879728e-03  7.54090643e-03 -1.10152787e-02 -4.04248367e-03
  -7.24537298e-04 -4.03502454e-02 -3.46059332e-02 -7.92174591e-04
  -2.78487445e-02  4.22124460e-03 -2.34118625e-03  4.68893202e-02
  -2.40964309e-02  6.58263426e-03  0.00000000e+00 -1.06404899e-02
  -6.57115036e-04  1.83736987e-03  4.71624881e-03  4.59326667e-03
  -2.11296436e-02  7.69920334e-03  9.63588835e-03  6.97926589e-03
  -1.31042500e-02  5.12822543e-04  1.13357609e-02  7.51542321e-04
  -2.96353118e-02 -1.27433769e-02  4.35726615e-03  6.12298723e-04
  -2.87901293e-02  8.37917698e-03  4.15168647e-03  6.05821477e-03
   1.69256363e-02 -3.96513268e-02  8.41897903e-03  2.59189050e-02
  -2.40119082e-02  3.51413642e-02 -1.20591670e-02  4.02558898e-03
   3.64187455e-02 -2.15658178e-02 -8.83495685e-03 -9.75622677e-05
  -1.07509087e-02  5.12346974e-03 -3.10714314e-02 -9.04470855e-03
  -1.59977282e-02 -1.46438349e-02 -1.78883469e-02 -1.92557780e-02
   1.78333800e-03 -1.03883290e-03  9.93805978e-04 -2.24903444e-02
   1.16720118e-07  5.36833867e-02  1.95375589e-03 -2.11893072e-02
   1.07674429e-02 -5.14852512e-03 -5.69404369e-02 -4.48608520e-02
   7.93865362e-02  1.16369602e-02  8.60992692e-03  4.13415067e-03
  -2.38109424e-02 -1.16553406e-02  2.12244496e-03 -3.24198574e-02
   1.85734204e-02 -6.67553656e-03  0.00000000e+00  8.93218329e-03
  -3.78716311e-02 -6.08784528e-03 -2.19712782e-02  3.89818898e-03
   5.78575117e-03  2.84085929e-03 -1.11183069e-02  4.26315805e-03
  -5.78328837e-03 -1.86083811e-02 -9.31472045e-03  1.71179464e-02
   2.78352707e-03 -5.24710885e-03 -1.58906177e-02 -2.57995935e-02
  -3.54469438e-03 -3.04936422e-03 -2.23340727e-02 -2.05275732e-03
  -1.51540270e-03 -2.89039785e-03  1.35720572e-02  5.00110540e-03
  -2.03782712e-03  1.31302093e-02  3.12155947e-03  9.66232590e-03
  -3.82160421e-03  1.59629542e-02 -3.13063947e-02 -7.70711547e-02
  -4.53873213e-02  2.44396713e-02  1.16176026e-02 -4.28616324e-03
   2.93571138e-03 -3.21378840e-03 -5.05168012e-03  2.14775046e-02
   9.30468055e-03  1.07742979e-02  1.09680076e-02  3.36540862e-03
   1.01536295e-02  2.41527039e-03 -5.70225641e-03  1.03883289e-02
   1.02343923e-03  1.32746365e-02 -8.20576467e-04 -6.66446767e-03
   7.90373582e-04  7.65092311e-03  1.32660069e-02  2.26848794e-02
   2.28302296e-03 -1.40102265e-02 -3.71336568e-03  2.64946628e-02
  -1.79877478e-02  1.81385643e-02  3.41749975e-02 -3.40065409e-03
   5.02072962e-03 -8.78754424e-04 -2.26040006e-03  1.29404119e-03
  -5.50349194e-03  2.05685501e-03 -2.00369264e-03  9.98966381e-03
  -3.18727710e-02  3.82714327e-05  3.85631484e-04  1.10011642e-02
  -2.09306138e-02 -1.51879536e-02 -1.86591424e-02 -4.32213442e-02
   1.26460455e-02  2.51820690e-02  7.44768353e-03  2.62360977e-02
  -3.11786477e-03  1.65680606e-02 -1.20075957e-02 -2.05131750e-03
   5.09738736e-03  3.23257670e-02 -3.96595964e-03  1.85930483e-02
  -2.01630691e-02  1.19437868e-02  1.96814717e-03  1.59577917e-02
  -1.34377019e-02  3.74845529e-04  1.97689620e-02 -8.72276622e-04
   2.34626536e-03 -2.57764641e-02 -1.16449587e-02  1.23092515e-02
   3.96490754e-02 -3.16921730e-02 -2.32013547e-02  6.75377040e-04
  -6.54763055e-04  9.46946754e-03  5.48372418e-04  7.70140168e-03
  -4.63439632e-02 -7.70876405e-04  2.54325016e-02  2.66350587e-02
   3.31780889e-02  0.00000000e+00 -1.35134870e-02  6.09026263e-04
   3.24610758e-03 -4.83750860e-03]]
intercept: [-0.29125527]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_3.pickle')
test_set.shape

(153993, 787)

# test classifier
predic = best_model.predict(test_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
print("predictions:\n", predic)

predictions:
 [0 1 0 ... 0 1 1]

set(predic)

{0, 1}

# evaluate classifier

print("Report for Support Vector Machine:")
print(_classification_report(test_set['likes'], predic))

print("Accuracy for Support Vector Machine:", _accuracy_score(test_set['likes'], predic)*100)

Report for Support Vector Machine:
              precision    recall  f1-score   support

           0       0.65      0.35      0.45     50930
           1       0.74      0.91      0.81    103063

    accuracy                           0.72    153993
   macro avg       0.70      0.63      0.63    153993
weighted avg       0.71      0.72      0.70    153993

Accuracy for Support Vector Machine: 72.28900014935743

# Confusion matrix for SVC

print("Confusion Matrix for SVC: ")
_confusion_matrix(test_set['likes'], predic)

Confusion Matrix for SVC:

array([[17698, 33232],
       [ 9441, 93622]], dtype=int64)

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], predic)

_plt.plot(fpr,tpr)
_plt.xlim([0.0,1.0])
_plt.ylim([0.0,1.0])

_plt.title("SVM ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_del_all()

7.5 Random forest without fake reviews¶

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')
train_set.shape

(558386, 787)

train_set = train_set[train_set['bin_truth_score']!=-1]
train_set.shape

(438571, 787)

params = _jl.load("../models/best_Random_Forest_2.joblib").get_params()
params['n_jobs'] = -1
params['verbose'] = 10
best_model = _RandomForestClassifier(**params)
best_model.get_params()

{'bootstrap': False,
 'class_weight': None,
 'criterion': 'entropy',
 'max_depth': 50,
 'max_features': 'auto',
 'max_leaf_nodes': None,
 'min_impurity_decrease': 0.0,
 'min_impurity_split': None,
 'min_samples_leaf': 2,
 'min_samples_split': 10,
 'min_weight_fraction_leaf': 0.0,
 'n_estimators': 1000,
 'n_jobs': -1,
 'oob_score': False,
 'random_state': None,
 'verbose': 10,
 'warm_start': False}

best_model.fit(train_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])

[Parallel(n_jobs=-1)]: Using backend ThreadingBackend with 12 concurrent workers.

building tree 1 of 1000
building tree 2 of 1000
building tree 3 of 1000
building tree 4 of 1000
building tree 5 of 1000
building tree 6 of 1000
building tree 7 of 1000
building tree 8 of 1000
building tree 9 of 1000
building tree 10 of 1000
building tree 11 of 1000
building tree 12 of 1000
building tree 13 of 1000

[Parallel(n_jobs=-1)]: Done   1 tasks      | elapsed:   16.4s

building tree 14 of 1000
building tree 15 of 1000
building tree 16 of 1000
building tree 17 of 1000
building tree 18 of 1000
building tree 19 of 1000
building tree 20 of 1000
building tree 21 of 1000

[Parallel(n_jobs=-1)]: Done   8 tasks      | elapsed:   20.2s

building tree 22 of 1000
building tree 23 of 1000
building tree 24 of 1000
building tree 25 of 1000
building tree 26 of 1000
building tree 27 of 1000
building tree 28 of 1000

[Parallel(n_jobs=-1)]: Done  17 tasks      | elapsed:   37.7s

building tree 29 of 1000
building tree 30 of 1000
building tree 31 of 1000
building tree 32 of 1000
building tree 33 of 1000
building tree 34 of 1000
building tree 35 of 1000
building tree 36 of 1000
building tree 37 of 1000

[Parallel(n_jobs=-1)]: Done  26 tasks      | elapsed:   55.7s

building tree 38 of 1000
building tree 39 of 1000
building tree 40 of 1000
building tree 41 of 1000
building tree 42 of 1000
building tree 43 of 1000
building tree 44 of 1000
building tree 45 of 1000
building tree 46 of 1000
building tree 47 of 1000
building tree 48 of 1000
building tree 49 of 1000
building tree 50 of 1000

_jl.dump(best_model, "../models/best_Random_Forest_fake.joblib")

['../models/best_Random_Forest_fake.joblib']

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_3.pickle')
test_set.shape

(153993, 787)

# test classifier
predic = best_model.predict(test_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
print("predictions:\n", predic)

[Parallel(n_jobs=12)]: Using backend ThreadingBackend with 12 concurrent workers.
[Parallel(n_jobs=12)]: Done   1 tasks      | elapsed:    0.1s
[Parallel(n_jobs=12)]: Done   8 tasks      | elapsed:    0.1s
[Parallel(n_jobs=12)]: Done  17 tasks      | elapsed:    0.4s
[Parallel(n_jobs=12)]: Done  26 tasks      | elapsed:    0.7s
[Parallel(n_jobs=12)]: Done  37 tasks      | elapsed:    1.0s
[Parallel(n_jobs=12)]: Done  48 tasks      | elapsed:    1.2s
[Parallel(n_jobs=12)]: Done  61 tasks      | elapsed:    1.6s
[Parallel(n_jobs=12)]: Done  74 tasks      | elapsed:    1.9s
[Parallel(n_jobs=12)]: Done  89 tasks      | elapsed:    2.2s
[Parallel(n_jobs=12)]: Done 104 tasks      | elapsed:    2.6s
[Parallel(n_jobs=12)]: Done 121 tasks      | elapsed:    3.0s
[Parallel(n_jobs=12)]: Done 138 tasks      | elapsed:    3.4s
[Parallel(n_jobs=12)]: Done 157 tasks      | elapsed:    3.9s
[Parallel(n_jobs=12)]: Done 176 tasks      | elapsed:    4.3s
[Parallel(n_jobs=12)]: Done 197 tasks      | elapsed:    4.9s
[Parallel(n_jobs=12)]: Done 218 tasks      | elapsed:    5.4s
[Parallel(n_jobs=12)]: Done 241 tasks      | elapsed:    6.0s
[Parallel(n_jobs=12)]: Done 264 tasks      | elapsed:    6.5s
[Parallel(n_jobs=12)]: Done 289 tasks      | elapsed:    7.2s
[Parallel(n_jobs=12)]: Done 314 tasks      | elapsed:    7.8s
[Parallel(n_jobs=12)]: Done 341 tasks      | elapsed:    8.5s
[Parallel(n_jobs=12)]: Done 368 tasks      | elapsed:    9.1s
[Parallel(n_jobs=12)]: Done 397 tasks      | elapsed:    9.8s
[Parallel(n_jobs=12)]: Done 426 tasks      | elapsed:   10.5s
[Parallel(n_jobs=12)]: Done 457 tasks      | elapsed:   11.3s
[Parallel(n_jobs=12)]: Done 488 tasks      | elapsed:   12.0s
[Parallel(n_jobs=12)]: Done 521 tasks      | elapsed:   12.9s
[Parallel(n_jobs=12)]: Done 554 tasks      | elapsed:   13.7s
[Parallel(n_jobs=12)]: Done 589 tasks      | elapsed:   14.6s
[Parallel(n_jobs=12)]: Done 624 tasks      | elapsed:   15.4s
[Parallel(n_jobs=12)]: Done 661 tasks      | elapsed:   16.3s
[Parallel(n_jobs=12)]: Done 698 tasks      | elapsed:   17.2s
[Parallel(n_jobs=12)]: Done 737 tasks      | elapsed:   18.2s
[Parallel(n_jobs=12)]: Done 776 tasks      | elapsed:   19.1s
[Parallel(n_jobs=12)]: Done 817 tasks      | elapsed:   20.1s
[Parallel(n_jobs=12)]: Done 858 tasks      | elapsed:   21.2s
[Parallel(n_jobs=12)]: Done 901 tasks      | elapsed:   22.2s
[Parallel(n_jobs=12)]: Done 944 tasks      | elapsed:   23.3s

predictions:
 [0 0 1 ... 0 0 0]

[Parallel(n_jobs=12)]: Done 1000 out of 1000 | elapsed:   24.5s finished

set(predic)

{0, 1}

# evaluate classifier

print("Report for Random Forest classifier:")
print(_classification_report(test_set['likes'], predic))

print("Accuracy for Random Forest Classifier:", _accuracy_score(test_set['likes'], predic)*100)

Report for Random Forest classifier:
              precision    recall  f1-score   support

           0       0.68      0.41      0.51     50930
           1       0.76      0.90      0.82    103063

    accuracy                           0.74    153993
   macro avg       0.72      0.66      0.67    153993
weighted avg       0.73      0.74      0.72    153993

Accuracy for Random Forest Classifier: 74.14947432675511

# Confusion matrix for Random Forest

print("Confusion Matrix for Random Forest: ")
_confusion_matrix(test_set['likes'], predic)

Confusion Matrix for Random Forest:

array([[21058, 29872],
       [ 9936, 93127]], dtype=int64)

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], predic)

_plt.plot(fpr,tpr)
_plt.xlim([0.0,1.0])
_plt.ylim([0.0,1.0])

_plt.title("Random Forest ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_del_all()

7.6 Neural Network without fake reviews¶

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')
train_set.shape

(558386, 787)

train_set = train_set[train_set['bin_truth_score']!=-1]
train_set.shape

(438571, 787)

# Number of neurons for hiddens layer
number_samples, number_features = train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']).shape

number_input_neurons = number_features
number_output_neurons = 1
number_train_semples = number_samples
alpha = 7

number_hidden_neurons = round(number_train_semples / (alpha * (number_input_neurons + number_output_neurons)))
print(number_hidden_neurons)

80

classifier = _Sequential()

# First Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal', input_dim = number_features))

# Second Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Third Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fourth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fifth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Output Layer
classifier.add(_Dense(1, activation = 'sigmoid', kernel_initializer = 'random_normal'))

# Compiling the neural network
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

# Fitting the data to the training dataset
classifier.fit(train_set.drop(
    columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']),
    train_set['likes'], validation_split = 0.3, batch_size = 100, epochs = 100)

Train on 306999 samples, validate on 131572 samples
Epoch 1/100
306999/306999 [==============================] - 30s 98us/step - loss: 0.6015 - acc: 0.6869 - val_loss: 0.5663 - val_acc: 0.7137
Epoch 2/100
306999/306999 [==============================] - 24s 79us/step - loss: 0.5644 - acc: 0.7154 - val_loss: 0.5518 - val_acc: 0.7288
Epoch 3/100
306999/306999 [==============================] - 24s 80us/step - loss: 0.5541 - acc: 0.7242 - val_loss: 0.5727 - val_acc: 0.7173
Epoch 4/100
306999/306999 [==============================] - 24s 77us/step - loss: 0.5498 - acc: 0.7276 - val_loss: 0.5402 - val_acc: 0.7353
Epoch 5/100
306999/306999 [==============================] - 25s 82us/step - loss: 0.5467 - acc: 0.7291 - val_loss: 0.5481 - val_acc: 0.7297
Epoch 6/100
306999/306999 [==============================] - 25s 80us/step - loss: 0.5438 - acc: 0.7325 - val_loss: 0.5427 - val_acc: 0.7352
Epoch 7/100
306999/306999 [==============================] - 24s 78us/step - loss: 0.5427 - acc: 0.7321 - val_loss: 0.5657 - val_acc: 0.7258
Epoch 8/100
306999/306999 [==============================] - 23s 74us/step - loss: 0.5409 - acc: 0.7333 - val_loss: 0.5452 - val_acc: 0.7307
Epoch 9/100
306999/306999 [==============================] - 25s 82us/step - loss: 0.5399 - acc: 0.7343 - val_loss: 0.5414 - val_acc: 0.7344
Epoch 10/100
306999/306999 [==============================] - 24s 77us/step - loss: 0.5380 - acc: 0.7360 - val_loss: 0.5394 - val_acc: 0.7367
Epoch 11/100
306999/306999 [==============================] - 23s 75us/step - loss: 0.5369 - acc: 0.7362 - val_loss: 0.5342 - val_acc: 0.7394
Epoch 12/100
306999/306999 [==============================] - 22s 73us/step - loss: 0.5357 - acc: 0.7378 - val_loss: 0.5382 - val_acc: 0.7376
Epoch 13/100
306999/306999 [==============================] - 22s 73us/step - loss: 0.5353 - acc: 0.7379 - val_loss: 0.5364 - val_acc: 0.7400
Epoch 14/100
306999/306999 [==============================] - 23s 74us/step - loss: 0.5338 - acc: 0.7385 - val_loss: 0.5361 - val_acc: 0.7408
Epoch 15/100
306999/306999 [==============================] - 22s 72us/step - loss: 0.5335 - acc: 0.7395 - val_loss: 0.5489 - val_acc: 0.7342
Epoch 16/100
306999/306999 [==============================] - 21s 69us/step - loss: 0.5319 - acc: 0.7397 - val_loss: 0.5329 - val_acc: 0.7412
Epoch 17/100
306999/306999 [==============================] - 20s 66us/step - loss: 0.5318 - acc: 0.7403 - val_loss: 0.5357 - val_acc: 0.7415
Epoch 18/100
306999/306999 [==============================] - 21s 67us/step - loss: 0.5303 - acc: 0.7404 - val_loss: 0.5333 - val_acc: 0.7393
Epoch 19/100
306999/306999 [==============================] - 21s 67us/step - loss: 0.5301 - acc: 0.7416 - val_loss: 0.5410 - val_acc: 0.7401
Epoch 20/100
306999/306999 [==============================] - 21s 67us/step - loss: 0.5303 - acc: 0.7412 - val_loss: 0.5495 - val_acc: 0.7352
Epoch 21/100
306999/306999 [==============================] - 21s 67us/step - loss: 0.5288 - acc: 0.7425 - val_loss: 0.5292 - val_acc: 0.7440
Epoch 22/100
306999/306999 [==============================] - 21s 67us/step - loss: 0.5287 - acc: 0.7426 - val_loss: 0.5360 - val_acc: 0.7387
Epoch 23/100
306999/306999 [==============================] - 21s 67us/step - loss: 0.5281 - acc: 0.7431 - val_loss: 0.5454 - val_acc: 0.7360
Epoch 24/100
306999/306999 [==============================] - 21s 68us/step - loss: 0.5266 - acc: 0.7439 - val_loss: 0.5505 - val_acc: 0.7377
Epoch 25/100
306999/306999 [==============================] - 21s 69us/step - loss: 0.5270 - acc: 0.7433 - val_loss: 0.5461 - val_acc: 0.7369
Epoch 26/100
306999/306999 [==============================] - 21s 69us/step - loss: 0.5266 - acc: 0.7432 - val_loss: 0.5731 - val_acc: 0.7101
Epoch 27/100
306999/306999 [==============================] - 21s 69us/step - loss: 0.5292 - acc: 0.7417 - val_loss: 0.5589 - val_acc: 0.7399
Epoch 28/100
306999/306999 [==============================] - 22s 70us/step - loss: 0.5257 - acc: 0.7439 - val_loss: 0.5932 - val_acc: 0.7362
Epoch 29/100
306999/306999 [==============================] - 22s 72us/step - loss: 0.5253 - acc: 0.7445 - val_loss: 0.5859 - val_acc: 0.7417
Epoch 30/100
306999/306999 [==============================] - 24s 78us/step - loss: 0.5246 - acc: 0.7448 - val_loss: 0.5809 - val_acc: 0.7397
Epoch 31/100
306999/306999 [==============================] - 22s 72us/step - loss: 0.5274 - acc: 0.7438 - val_loss: 0.5833 - val_acc: 0.7371
Epoch 32/100
306999/306999 [==============================] - 22s 72us/step - loss: 0.5249 - acc: 0.7450 - val_loss: 0.5755 - val_acc: 0.7412
Epoch 33/100
306999/306999 [==============================] - 22s 72us/step - loss: 0.5246 - acc: 0.7448 - val_loss: 0.5392 - val_acc: 0.7373
Epoch 34/100
306999/306999 [==============================] - 22s 73us/step - loss: 0.5240 - acc: 0.7453 - val_loss: 0.5380 - val_acc: 0.7442
Epoch 35/100
306999/306999 [==============================] - 22s 72us/step - loss: 0.5233 - acc: 0.7457 - val_loss: 0.5578 - val_acc: 0.7397
Epoch 36/100
306999/306999 [==============================] - 22s 71us/step - loss: 0.5230 - acc: 0.7452 - val_loss: 0.5813 - val_acc: 0.7399
Epoch 37/100
306999/306999 [==============================] - 22s 70us/step - loss: 0.5224 - acc: 0.7457 - val_loss: 0.5773 - val_acc: 0.7428
Epoch 38/100
306999/306999 [==============================] - 22s 71us/step - loss: 0.5218 - acc: 0.7462 - val_loss: 0.5826 - val_acc: 0.7411
Epoch 39/100
306999/306999 [==============================] - 22s 70us/step - loss: 0.5226 - acc: 0.7463 - val_loss: 0.5797 - val_acc: 0.7433
Epoch 40/100
306999/306999 [==============================] - 22s 70us/step - loss: 0.5215 - acc: 0.7471 - val_loss: 0.5719 - val_acc: 0.7411
Epoch 41/100
306999/306999 [==============================] - 22s 70us/step - loss: 0.5230 - acc: 0.7455 - val_loss: 0.5542 - val_acc: 0.7255
Epoch 42/100
306999/306999 [==============================] - 21s 70us/step - loss: 0.5279 - acc: 0.7414 - val_loss: 0.5689 - val_acc: 0.7371
Epoch 43/100
306999/306999 [==============================] - 22s 71us/step - loss: 0.5207 - acc: 0.7465 - val_loss: 0.5660 - val_acc: 0.7420
Epoch 44/100
306999/306999 [==============================] - 21s 70us/step - loss: 0.5209 - acc: 0.7468 - val_loss: 0.5628 - val_acc: 0.7442
Epoch 45/100
306999/306999 [==============================] - 20s 67us/step - loss: 0.5210 - acc: 0.7471 - val_loss: 0.5751 - val_acc: 0.7384
Epoch 46/100
306999/306999 [==============================] - 20s 66us/step - loss: 0.5202 - acc: 0.7466 - val_loss: 0.5986 - val_acc: 0.7347
Epoch 47/100
306999/306999 [==============================] - 20s 66us/step - loss: 0.5201 - acc: 0.7471 - val_loss: 0.5670 - val_acc: 0.7426
Epoch 48/100
306999/306999 [==============================] - 20s 66us/step - loss: 0.5199 - acc: 0.7479 - val_loss: 0.5761 - val_acc: 0.7436
Epoch 49/100
306999/306999 [==============================] - 20s 66us/step - loss: 0.5192 - acc: 0.7479 - val_loss: 0.5831 - val_acc: 0.7419
Epoch 50/100
306999/306999 [==============================] - 20s 66us/step - loss: 0.5191 - acc: 0.7474 - val_loss: 0.5882 - val_acc: 0.7405
Epoch 51/100
306999/306999 [==============================] - 20s 67us/step - loss: 0.5198 - acc: 0.7475 - val_loss: 0.5773 - val_acc: 0.7406
Epoch 52/100
306999/306999 [==============================] - 20s 67us/step - loss: 0.5188 - acc: 0.7484 - val_loss: 0.5844 - val_acc: 0.7380
Epoch 53/100
306999/306999 [==============================] - 20s 66us/step - loss: 0.5183 - acc: 0.7494 - val_loss: 0.5901 - val_acc: 0.7401
Epoch 54/100
306999/306999 [==============================] - 20s 67us/step - loss: 0.5185 - acc: 0.7483 - val_loss: 0.5839 - val_acc: 0.7434
Epoch 55/100
306999/306999 [==============================] - 20s 66us/step - loss: 0.5180 - acc: 0.7492 - val_loss: 0.5861 - val_acc: 0.7441
Epoch 56/100
306999/306999 [==============================] - 20s 66us/step - loss: 0.5186 - acc: 0.7481 - val_loss: 0.5783 - val_acc: 0.7400
Epoch 57/100
306999/306999 [==============================] - 20s 65us/step - loss: 0.5187 - acc: 0.7484 - val_loss: 0.5845 - val_acc: 0.7431
Epoch 58/100
306999/306999 [==============================] - 20s 64us/step - loss: 0.5180 - acc: 0.7490 - val_loss: 0.6011 - val_acc: 0.7428
Epoch 59/100
306999/306999 [==============================] - 20s 64us/step - loss: 0.5171 - acc: 0.7492 - val_loss: 0.5731 - val_acc: 0.7454
Epoch 60/100
306999/306999 [==============================] - 20s 64us/step - loss: 0.5187 - acc: 0.7484 - val_loss: 0.6184 - val_acc: 0.7342
Epoch 61/100
306999/306999 [==============================] - 20s 65us/step - loss: 0.5175 - acc: 0.7487 - val_loss: 0.6044 - val_acc: 0.7372
Epoch 62/100
306999/306999 [==============================] - 20s 65us/step - loss: 0.5175 - acc: 0.7491 - val_loss: 0.5877 - val_acc: 0.7409
Epoch 63/100
306999/306999 [==============================] - 20s 64us/step - loss: 0.5161 - acc: 0.7503 - val_loss: 0.5970 - val_acc: 0.7435
Epoch 64/100
306999/306999 [==============================] - 20s 64us/step - loss: 0.5164 - acc: 0.7499 - val_loss: 0.6022 - val_acc: 0.7432
Epoch 65/100
306999/306999 [==============================] - 20s 64us/step - loss: 0.5177 - acc: 0.7486 - val_loss: 0.5970 - val_acc: 0.7408
Epoch 66/100
306999/306999 [==============================] - 20s 64us/step - loss: 0.5163 - acc: 0.7497 - val_loss: 0.6019 - val_acc: 0.7426
Epoch 67/100
306999/306999 [==============================] - 20s 65us/step - loss: 0.5160 - acc: 0.7499 - val_loss: 0.5921 - val_acc: 0.7435
Epoch 68/100
306999/306999 [==============================] - 20s 64us/step - loss: 0.5158 - acc: 0.7501 - val_loss: 0.5989 - val_acc: 0.7414
Epoch 69/100
306999/306999 [==============================] - 20s 65us/step - loss: 0.5160 - acc: 0.7502 - val_loss: 0.5861 - val_acc: 0.7439
Epoch 70/100
306999/306999 [==============================] - 20s 65us/step - loss: 0.5155 - acc: 0.7502 - val_loss: 0.5820 - val_acc: 0.7433
Epoch 71/100
306999/306999 [==============================] - 20s 65us/step - loss: 0.5156 - acc: 0.7501 - val_loss: 0.5895 - val_acc: 0.7413
Epoch 72/100
306999/306999 [==============================] - 20s 64us/step - loss: 0.5155 - acc: 0.7500 - val_loss: 0.5861 - val_acc: 0.7421
Epoch 73/100
306999/306999 [==============================] - 20s 64us/step - loss: 0.5150 - acc: 0.7513 - val_loss: 0.5808 - val_acc: 0.7405
Epoch 74/100
306999/306999 [==============================] - 21s 68us/step - loss: 0.5152 - acc: 0.7508 - val_loss: 0.5894 - val_acc: 0.7420
Epoch 75/100
306999/306999 [==============================] - 20s 65us/step - loss: 0.5145 - acc: 0.7511 - val_loss: 0.6106 - val_acc: 0.7419
Epoch 76/100
306999/306999 [==============================] - 20s 64us/step - loss: 0.5137 - acc: 0.7517 - val_loss: 0.5926 - val_acc: 0.7430
Epoch 77/100
306999/306999 [==============================] - 21s 67us/step - loss: 0.5140 - acc: 0.7510 - val_loss: 0.5935 - val_acc: 0.7418
Epoch 78/100
306999/306999 [==============================] - 22s 73us/step - loss: 0.5136 - acc: 0.7508 - val_loss: 0.5996 - val_acc: 0.7415
Epoch 79/100
306999/306999 [==============================] - 23s 74us/step - loss: 0.5137 - acc: 0.7514 - val_loss: 0.6083 - val_acc: 0.7355
Epoch 80/100
306999/306999 [==============================] - 24s 77us/step - loss: 0.5136 - acc: 0.7511 - val_loss: 0.6018 - val_acc: 0.7445
Epoch 81/100
306999/306999 [==============================] - 24s 77us/step - loss: 0.5134 - acc: 0.7516 - val_loss: 0.5918 - val_acc: 0.7459
Epoch 82/100
306999/306999 [==============================] - 22s 70us/step - loss: 0.5132 - acc: 0.7519 - val_loss: 0.5895 - val_acc: 0.7433
Epoch 83/100
306999/306999 [==============================] - 24s 79us/step - loss: 0.5129 - acc: 0.7516 - val_loss: 0.6002 - val_acc: 0.7399
Epoch 84/100
306999/306999 [==============================] - 24s 79us/step - loss: 0.5133 - acc: 0.7516 - val_loss: 0.6034 - val_acc: 0.7421
Epoch 85/100
306999/306999 [==============================] - 20s 67us/step - loss: 0.5130 - acc: 0.7516 - val_loss: 0.5826 - val_acc: 0.7426
Epoch 86/100
306999/306999 [==============================] - 20s 66us/step - loss: 0.5130 - acc: 0.7520 - val_loss: 0.5880 - val_acc: 0.7458
Epoch 87/100
306999/306999 [==============================] - 20s 65us/step - loss: 0.5125 - acc: 0.7524 - val_loss: 0.5789 - val_acc: 0.7446
Epoch 88/100
306999/306999 [==============================] - 20s 65us/step - loss: 0.5127 - acc: 0.7517 - val_loss: 0.5776 - val_acc: 0.7436
Epoch 89/100
306999/306999 [==============================] - 21s 68us/step - loss: 0.5120 - acc: 0.7520 - val_loss: 0.5866 - val_acc: 0.7444
Epoch 90/100
306999/306999 [==============================] - 21s 68us/step - loss: 0.5134 - acc: 0.7515 - val_loss: 0.5776 - val_acc: 0.7462
Epoch 91/100
306999/306999 [==============================] - 20s 65us/step - loss: 0.5118 - acc: 0.7524 - val_loss: 0.5934 - val_acc: 0.7398
Epoch 92/100
306999/306999 [==============================] - 20s 65us/step - loss: 0.5120 - acc: 0.7526 - val_loss: 0.5861 - val_acc: 0.7444
Epoch 93/100
306999/306999 [==============================] - 20s 66us/step - loss: 0.5121 - acc: 0.7516 - val_loss: 0.5867 - val_acc: 0.7470
Epoch 94/100
306999/306999 [==============================] - 20s 67us/step - loss: 0.5119 - acc: 0.7526 - val_loss: 0.5952 - val_acc: 0.7380
Epoch 95/100
306999/306999 [==============================] - 20s 65us/step - loss: 0.5118 - acc: 0.7521 - val_loss: 0.6002 - val_acc: 0.7436
Epoch 96/100
306999/306999 [==============================] - 20s 65us/step - loss: 0.5115 - acc: 0.7528 - val_loss: 0.5829 - val_acc: 0.7447
Epoch 97/100
306999/306999 [==============================] - 20s 66us/step - loss: 0.5111 - acc: 0.7534 - val_loss: 0.5996 - val_acc: 0.7415
Epoch 98/100
306999/306999 [==============================] - 23s 76us/step - loss: 0.5107 - acc: 0.7528 - val_loss: 0.5674 - val_acc: 0.7466
Epoch 99/100
306999/306999 [==============================] - 23s 74us/step - loss: 0.5108 - acc: 0.7530 - val_loss: 0.5815 - val_acc: 0.7454
Epoch 100/100
306999/306999 [==============================] - 23s 74us/step - loss: 0.5106 - acc: 0.7529 - val_loss: 0.5847 - val_acc: 0.7446

<keras.callbacks.History at 0x1ad8c4c5b00>

# Save trained model
classifier.save("../models/trained_deep_neural_network_fake.h5")

# evaluation of the model training
evaluation = classifier.evaluate(train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])
print(evaluation)

438571/438571 [==============================] - 12s 28us/step
[0.5348048495635159, 0.7506811895911467]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_3.pickle')
test_set.shape

(153993, 787)

prediction = classifier.predict(test_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
binary_prediction = _binarize(prediction, threshold = 0.5)
binary_prediction

array([[1.],
       [0.],
       [1.],
       ...,
       [0.],
       [1.],
       [1.]], dtype=float32)

set(binary_prediction[:,0])

{0.0, 1.0}

print("Report for Deep Learning approach:")
print(_classification_report(test_set['likes'], binary_prediction))

print("Accuracy for Deep Learning approach:", _accuracy_score(test_set['likes'], binary_prediction) * 100)

Report for Deep Learning approach:
              precision    recall  f1-score   support

           0       0.69      0.39      0.50     50930
           1       0.75      0.92      0.83    103063

    accuracy                           0.74    153993
   macro avg       0.72      0.65      0.66    153993
weighted avg       0.73      0.74      0.72    153993

Accuracy for Deep Learning approach: 74.06700304559298

matrix = _confusion_matrix(test_set['likes'], binary_prediction)
print(matrix)

[[19674 31256]
 [ 8679 94384]]

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], binary_prediction)

_plt.plot(fpr, tpr)
_plt.xlim([0.0, 1.0])
_plt.ylim([0.0, 1.0])

_plt.title("Neural Network ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_del_all()

7.7 SVM only standard columns¶

# columns we are going to remove from the dataset

_cols_std = ['cuisine_av_hist', 'coll_score', 'average_stars_review', 'num_reviews_review',
            'average_stars_user', 'num_reviews_user',
            'av_rat_chinese_cuisine', 'av_rat_japanese_cuisine', 'av_rat_mexican_cuisine', 'av_rat_italian_cuisine', 
            'av_rat_others_cuisine', 'av_rat_american_cuisine', 'av_rat_korean_cuisine', 'av_rat_mediterranean_cuisine',
            'av_rat_thai_cuisine', 'av_rat_asianfusion_cuisine']


_cols_bin = ['cuisine_av_hist_bin', 'bin_truth_score', 'coll_score_bin', 'average_stars_bin_review',
            'num_reviews_bin_review', 'average_stars_bin_user', 'num_reviews_bin_user',
            'av_rat_chinese_cuisine_bin', 'av_rat_japanese_cuisine_bin', 'av_rat_mexican_cuisine_bin', 
            'av_rat_italian_cuisine_bin', 'av_rat_others_cuisine_bin', 'av_rat_american_cuisine_bin', 
            'av_rat_korean_cuisine_bin', 'av_rat_mediterranean_cuisine_bin', 'av_rat_thai_cuisine_bin', 
            'av_rat_asianfusion_cuisine_bin']


_cols_real = ['cuisine_av_hist_real', 'real_truth_score', 'coll_score_real', 'average_stars_real_review',
             'num_reviews_real_review', 'average_stars_real_user', 'num_reviews_real_user',
             'av_rat_chinese_cuisine_real', 'av_rat_japanese_cuisine_real', 'av_rat_mexican_cuisine_real', 
             'av_rat_italian_cuisine_real', 'av_rat_others_cuisine_real', 'av_rat_american_cuisine_real', 
             'av_rat_korean_cuisine_real', 'av_rat_mediterranean_cuisine_real', 'av_rat_thai_cuisine_real', 
             'av_rat_asianfusion_cuisine_real']

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')
train_set.shape

(558386, 787)

train_set = train_set.drop(columns=[*_cols_bin, *_cols_real])
train_set.shape

(558386, 753)

best_model = _jl.load("../models/best_SVM.joblib")
best_model.set_params(verbose=10)
best_model.get_params()

{'C': 0.001,
 'class_weight': None,
 'dual': True,
 'fit_intercept': True,
 'intercept_scaling': 1,
 'loss': 'squared_hinge',
 'max_iter': 50000,
 'multi_class': 'ovr',
 'penalty': 'l2',
 'random_state': 0,
 'tol': 0.0001,
 'verbose': 10}

best_model.fit(train_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])

[LibLinear]

C:\Users\super\Anaconda3\envs\tf-gpu\lib\site-packages\sklearn\svm\base.py:929: ConvergenceWarning: Liblinear failed to converge, increase the number of iterations.
  "the number of iterations.", ConvergenceWarning)

LinearSVC(C=0.001, class_weight=None, dual=True, fit_intercept=True,
          intercept_scaling=1, loss='squared_hinge', max_iter=50000,
          multi_class='ovr', penalty='l2', random_state=0, tol=0.0001,
          verbose=10)

_jl.dump(best_model, "../models/best_SVM_std.joblib")

['../models/best_SVM_std.joblib']

print("coef:", best_model.coef_)
print("intercept:", best_model.intercept_)

coef: [[-1.45163681e-01 -2.06192678e-01  3.43101354e-01  9.50706829e-02
  -7.67267815e-02  1.75285027e-05 -3.59168241e-02  7.69028847e-03
  -3.37551424e-05  6.44365981e-01 -8.19044376e-05  9.98650771e-04
   2.85675869e-04 -8.07039934e-04 -5.88016148e-04 -6.42377233e-04
   1.10320520e-03 -2.13229419e-04 -1.36944603e-04  4.43540753e-04
  -4.35994723e-05  6.60548107e-05 -9.40240234e-06 -3.16915159e-04
  -1.36081275e-01  7.99684151e-05 -4.15239339e-02  6.73184431e-02
   3.00972016e-04  1.54171072e-02 -5.04502256e-04 -2.20501470e-04
   4.92195916e-05  1.81076061e-04  3.69069303e-05 -5.94465246e-04
   3.11871960e-02  2.33957370e-02  3.60430435e-02  4.00775537e-02
   5.32107117e-02  4.93676643e-02  1.37202769e-02  2.57261458e-02
   1.96609415e-02  4.11802748e-02 -1.30191311e-01 -9.69832815e-02
  -1.33655698e-01 -8.26716794e-02 -1.30493220e-01 -1.47665391e-01
  -1.32102583e-01 -1.12734326e-01 -1.15993381e-01 -1.25985560e-01
  -1.18534497e-01 -1.16310232e-01 -8.86037761e-02 -8.85625045e-02
  -9.81470552e-02 -8.55169544e-02 -9.25029457e-02 -9.31002298e-02
  -8.93448873e-02 -8.58822273e-02  6.08734051e-03  2.05795751e-02
   1.41080355e-02  2.39124916e-02  2.52071186e-02 -2.27900951e-02
   1.23261655e-02  3.62203871e-02 -5.55927439e-02  4.58146366e-02
   2.12876520e-02 -8.06569307e-02 -1.36249406e-02  3.25710384e-02
   2.78837538e-02  6.69934088e-02 -8.48508124e-03 -7.87466725e-02
  -1.35055624e-02 -3.24469542e-02 -3.84007855e-04 -1.56273648e-03
   1.46397663e-02  2.61100375e-02  4.14629939e-03  6.17941398e-03
   3.62395677e-01 -1.08990669e-02  1.53841378e-02 -9.16119301e-03
   1.96047295e-02  1.47209564e-02  4.01644957e-02 -2.06227493e-02
   2.23082188e-02 -1.90273869e-01  2.03107121e-02  5.93510277e-02
  -2.42890723e-02 -1.20876657e-01  6.00448898e-04 -9.27465261e-02
   2.47100759e-02 -2.73764890e-02 -5.80382982e-02  1.49861963e-02
   8.21008906e-03  1.22036178e-02 -8.81020521e-03  1.94235319e-02
  -6.96733312e-02  1.70162579e-02  1.91164345e-02  7.70633647e-03
   3.39672283e-02 -1.30166942e-01  5.28450257e-03 -1.07200769e-01
  -2.36470907e-02  2.43130012e-03 -5.27724141e-03 -1.38757559e-02
   7.08733692e-02  2.04349746e-02  3.16352746e-02  2.21865642e-02
  -1.02251176e-01  2.05783657e-02 -3.61030277e-02 -8.15191661e-02
   1.20569269e-02 -4.44393092e-02  3.24087647e-02 -1.95265101e-02
   6.17565705e-02  9.43867895e-02  1.74285353e-02  1.07599325e-02
   1.30084458e-02 -8.81609605e-03  3.87091663e-03 -4.50570854e-03
   1.50290373e-02  1.30466315e-01 -1.71555597e-02 -1.94424266e-02
  -1.41829606e-01 -1.03278840e-02 -2.79446846e-02  7.84417499e-03
   4.21330297e-02  7.30523616e-02 -2.78513184e-02  2.27285676e-02
   1.19320119e-01 -2.31984596e-02 -3.93233659e-02  3.08146120e-02
  -2.19267468e-02 -3.34173608e-02 -4.12674350e-03  5.30846615e-02
  -1.02563100e-02 -2.41766092e-02 -8.59153714e-02  1.26192469e-02
  -2.29300876e-02 -1.49246593e-02  8.37885241e-02 -1.87616542e-02
   2.35727338e-02  3.30342140e-03 -8.60404399e-02  3.09988524e-02
  -9.38913252e-03 -6.27554430e-02  1.23030711e-02 -8.10424577e-03
  -4.08190234e-03 -5.24827915e-03 -3.18628761e-03 -2.44255827e-02
   7.19578375e-02 -1.68246204e-02 -9.91665265e-03  1.83777613e-02
  -1.17397590e-01 -1.14693799e-02  1.21929127e-02  5.62237299e-03
  -1.46270968e-02 -9.23993972e-02  1.09020562e-02  1.86699249e-02
  -5.25853090e-02 -1.55957411e-02 -2.21546449e-04 -2.97786738e-02
   4.86462359e-02 -1.55229633e-02  4.18373542e-02  1.71467714e-02
  -5.19918848e-03  3.81667301e-02  3.48579483e-02 -4.92319566e-03
  -1.29767823e-02  4.92732989e-02  1.42219888e-02 -5.58502934e-02
  -3.24365777e-02 -3.71407945e-02 -3.41474181e-02 -9.26607323e-02
   6.07485058e-03 -7.19846901e-03 -3.02702141e-03 -8.06356475e-04
   2.36656810e-02  4.61743673e-03 -4.37871909e-02 -1.06015916e-02
   7.05669008e-02  1.75549103e-02  2.55293636e-02 -4.88283329e-02
  -4.75391122e-02  1.10785048e-02 -2.55689685e-02  6.79462775e-03
   4.92229083e-03  1.00597326e-02 -1.71726454e-02  3.02456881e-02
  -9.38494730e-03 -5.12942532e-02 -7.08734405e-03  1.65447502e-03
   1.13963471e-02 -3.87377211e-02 -2.93049321e-02 -1.51343551e-02
  -9.70864886e-04 -5.13886007e-02  8.47746494e-03  1.51853308e-02
   1.56862454e-02  1.48857768e-02  1.80193756e-02  1.56840457e-03
  -2.38286239e-03 -1.94318639e-02 -3.63279821e-03 -5.64898990e-03
   5.48698410e-03 -6.52145520e-03 -5.24582661e-02 -1.61122408e-03
  -4.09840236e-03  5.38503812e-03 -1.17971742e-02 -6.24407200e-02
  -6.15112432e-03  4.65114142e-02 -8.33013172e-03  3.04207318e-03
  -6.59838666e-02  1.26992765e-02  2.37586572e-03  4.03168509e-03
   2.63876042e-02  6.33459246e-03 -6.19889501e-03 -3.06410129e-02
  -5.82770709e-04 -1.21237923e-04  1.17629032e-02  1.74572679e-02
  -1.11831034e-02 -4.79644760e-02  9.79850843e-05  1.86125493e-02
   3.60662642e-03  2.21099674e-02 -3.18531006e-03  1.43292043e-02
  -1.58394074e-03 -4.88816456e-02  8.63630483e-03 -2.09455724e-02
   3.00491599e-02 -4.21879018e-02 -2.56323606e-02 -2.35893622e-03
   1.02321914e-02  2.15960707e-02 -2.22510931e-02 -2.32975547e-03
  -8.93205142e-02 -9.75105846e-02  5.86765283e-03  3.52497058e-03
  -5.26786065e-03  2.63459474e-02 -1.32625847e-02 -2.53487952e-03
   2.59372328e-03 -3.62331661e-02  5.27731603e-03 -1.67180664e-02
   3.75336453e-02  1.18794502e-02  4.98211223e-02 -1.39097098e-02
   1.69137552e-02  4.86807534e-02  6.35274356e-03  1.72525115e-02
  -7.02980518e-03  1.89758412e-02  8.32277890e-03  3.49231929e-03
   2.83064875e-03  6.27399920e-03  5.76510574e-02  1.62525003e-02
   1.99399239e-03  2.56095347e-03  1.79334613e-02 -2.02289848e-02
   2.24248422e-02  1.31479095e-02 -4.51964555e-03  1.34393935e-03
   8.20165971e-03  2.55749807e-03 -1.04731249e-03 -2.73817028e-02
   1.54034960e-02  7.17990583e-03  9.32720970e-03 -1.11028796e-02
  -1.14003566e-02 -4.53319447e-02 -1.15513923e-02  2.18462490e-02
   4.37686759e-02 -9.88918019e-03 -2.21166388e-03 -3.69104028e-02
   7.80945531e-03  1.19437844e-02 -3.57931051e-03  1.55373814e-02
  -7.98217588e-04  4.10792184e-03  3.21385425e-02  1.90861394e-02
   7.48471078e-03  2.09008427e-02 -3.20058693e-04  2.38828155e-02
  -1.60324453e-02  4.49793755e-03 -5.75234534e-02 -7.99101290e-03
  -2.09985879e-03 -1.27445527e-02 -1.31276495e-02  1.26768567e-02
   1.04755125e-02 -1.28173221e-02 -1.11561951e-02  2.10796938e-02
  -1.81560664e-02 -1.68085105e-02 -4.23834956e-02  1.03133064e-04
  -1.75092711e-03 -3.77780684e-03  1.76045984e-03 -1.04718589e-02
  -2.94780620e-02  1.46228167e-02 -2.13851279e-02  1.62646547e-02
  -3.62533898e-02 -2.79248386e-02  2.60336894e-02 -2.96256302e-02
   3.64294711e-02  4.01782015e-02 -1.15381863e-02 -9.34758284e-03
   3.02636863e-03 -4.61973800e-02 -1.43106402e-02  1.26470749e-02
  -1.79250218e-02  1.36500911e-02  6.55461631e-02 -8.78659972e-03
   2.05077333e-02 -3.16895781e-02  2.11768839e-03 -1.02883030e-03
   2.91978692e-03 -7.71149534e-04  8.08574147e-03  3.03699100e-02
  -5.42914411e-03 -2.66009296e-02 -7.69779587e-03 -6.34711234e-03
   2.06794028e-02  1.60466811e-02 -1.21135486e-02 -2.93338309e-03
  -2.53805631e-03 -2.97586051e-02  3.76067664e-04  4.76061734e-02
   1.82687395e-02 -1.47586696e-02 -1.72434028e-02  2.25708857e-02
   1.09903351e-02  1.79556823e-02 -2.06019069e-03  8.72050733e-03
   4.03168509e-03 -1.83472977e-03 -3.23866255e-03 -1.71672810e-04
   8.85174742e-03 -2.48307116e-02  2.26861094e-02  1.86370941e-03
  -5.40835197e-03  1.53280586e-02 -5.87940610e-02 -1.57940294e-02
   2.42943861e-02 -3.52404843e-02 -3.44167363e-02  6.25969029e-03
   7.36495219e-02 -6.52754991e-03  2.12075097e-02  1.23134272e-02
   2.33817581e-02  3.06627292e-02  7.93489692e-02  9.82910173e-03
   7.39245563e-03  1.20416792e-02  5.94771997e-03  1.10721685e-03
   5.56443605e-04 -4.19782527e-03 -5.23808454e-03 -1.13613990e-02
   2.73392763e-02 -2.39489952e-02  1.87873350e-02  1.13184848e-02
  -7.47975279e-03  3.60163882e-02 -2.47352058e-03 -2.73057798e-02
   1.19977320e-02 -1.55003681e-03  1.97790657e-02 -1.41076291e-02
  -8.15529979e-03  9.45926853e-03 -5.89448011e-03  2.15895408e-02
  -2.87217768e-02 -4.17266662e-02  5.28080035e-02  1.69717922e-03
  -2.82523414e-02  2.83681481e-02 -3.16756834e-02  1.62397233e-03
  -2.87693937e-02  2.39428680e-02 -2.19990863e-02  2.81463295e-03
   1.76802144e-02 -1.44572408e-02 -1.12283080e-03  7.26220147e-03
  -1.66535437e-03  2.05374763e-02 -3.53192241e-02 -1.88999403e-02
  -7.54961289e-03  1.46910010e-02  3.47818228e-02  3.98925887e-03
  -3.55625057e-02  2.31097459e-02 -4.34346282e-03 -1.32456198e-02
  -5.10248277e-02  4.56924066e-02 -4.85704689e-03  2.26144912e-02
   9.84353236e-03 -1.05166838e-02  2.32399374e-03 -3.03348354e-02
   2.93245043e-04 -4.65186267e-02  2.42918512e-02  5.39311475e-03
   1.20426883e-02  2.30813009e-02  2.62066536e-02  1.25079118e-02
  -2.12882894e-03  2.02904988e-02 -1.31141925e-02  2.37712078e-02
  -4.68004937e-02 -9.63250959e-04  8.29524536e-04 -1.46294253e-02
  -3.42737072e-02  3.02323993e-02  7.60810040e-03  4.83546051e-03
  -3.85942774e-02  1.70927471e-02  5.32095535e-03  1.24478457e-02
  -7.69370484e-03  1.60844295e-03  1.65410658e-03  5.38865552e-03
  -1.93543373e-02 -1.89373202e-03 -3.53189618e-04 -3.95985864e-02
  -3.19007089e-02 -9.42702217e-04 -3.43002747e-02  8.29840710e-03
  -2.31106105e-03  3.05407410e-02 -3.09533734e-02  6.54975585e-03
   0.00000000e+00 -1.35699369e-02  2.26533499e-03 -8.23491702e-03
   3.07512672e-03  3.16887105e-03 -2.64596838e-02 -5.75086219e-03
   5.26037966e-03  8.10317374e-03 -2.01879255e-02  5.88776060e-03
   9.42822804e-03 -1.11042564e-03 -4.18525910e-02 -1.05065751e-02
  -5.65779070e-03 -3.04874251e-03 -3.15533008e-02  8.28457531e-03
   2.23825082e-02  5.43544811e-03  2.19812500e-02 -3.59679807e-02
  -5.93539652e-03  2.86454153e-02 -1.40148849e-02  2.78413217e-02
  -4.77900579e-03  5.98548226e-03  3.53573210e-02 -3.65877840e-02
   1.96629870e-04 -1.18453721e-02 -1.34378108e-02 -3.01776304e-03
  -2.69667338e-02 -1.30932907e-03 -2.08677777e-02 -4.97819246e-03
  -2.13434891e-02 -1.19103945e-02  8.72717669e-03 -1.00409073e-03
   2.14271770e-03 -2.84418287e-02  5.71533918e-04  4.74604922e-02
   1.79895276e-03 -1.30045547e-02  5.43018082e-03 -6.92303704e-03
  -5.00501592e-02 -5.57593196e-02  7.84015193e-02  1.42319373e-02
   2.09516675e-02 -2.17050750e-03 -1.48057402e-02 -4.37628362e-03
   3.64140296e-03 -3.13656581e-02  1.23883263e-02 -1.11975056e-02
   0.00000000e+00  3.36492141e-03 -3.85754811e-02 -3.24125403e-03
  -3.35970132e-02 -4.26739329e-03  7.26938637e-03  3.34811871e-03
  -8.42574043e-03  2.38519007e-03 -5.86873734e-03 -3.20123998e-02
  -1.37741663e-02  1.01872997e-02  2.61936145e-03 -7.60006872e-04
  -2.10422733e-02 -2.67909105e-02 -1.30986999e-03 -1.64830501e-02
  -3.30349047e-02 -4.43917466e-03 -6.25434683e-03 -3.80937199e-04
   1.72541077e-03  6.82243224e-03  3.84294714e-04  1.42286898e-02
   3.10975022e-03  5.38613989e-03 -1.90145442e-03  4.15426267e-03
  -3.16312678e-02 -6.14377429e-02 -4.88267373e-02  3.21877339e-02
   4.76579631e-03  7.07853373e-05  9.87813003e-04  2.00657189e-03
  -2.68148319e-03  1.64774161e-02  2.23684921e-03 -4.95871863e-03
   9.93886413e-03  6.56974139e-03  1.04857756e-02  3.50319806e-03
  -4.74414277e-03  6.23655750e-04  6.68540992e-03  3.62368959e-02
  -5.04601765e-03 -1.27860877e-02 -4.53196564e-03  1.33891206e-02
   9.05215933e-03  1.65056530e-02  4.46993948e-03 -7.02486665e-03
  -1.05628404e-02  2.56795794e-02  1.64333464e-03  2.58463523e-02
   3.18329206e-02 -6.48480101e-04  1.36059796e-02 -1.58513158e-02
  -1.98690304e-03  1.08848003e-02 -4.16122045e-03  8.62606662e-03
  -1.40543879e-03  1.02193609e-02 -3.12528080e-02  2.11443671e-03
   4.22582619e-05  7.78117494e-03 -2.03078324e-02 -2.35145766e-02
  -1.48804847e-02 -3.76437508e-02  1.62879908e-02  2.54061169e-02
   7.23435225e-03  2.19549562e-02  1.16170832e-02  2.49230240e-02
  -2.29370719e-02 -7.58959791e-03 -3.46521558e-03  2.90983356e-02
  -1.09027779e-03  2.17484259e-02 -1.40010872e-02  9.87666984e-03
   1.71227705e-03  1.70408978e-02 -1.66326212e-02 -2.45024234e-03
   3.42578814e-02 -8.91890662e-04  9.07247035e-03 -1.96779982e-02
  -2.74027447e-03  1.24231076e-02  3.84661077e-02 -2.49017355e-02
  -2.72390968e-02  9.47283420e-04  1.19708791e-03  1.70745052e-02
   4.77623195e-04  7.92084136e-03 -4.39068101e-02 -4.61390943e-03
   1.89044803e-02  2.34377255e-02  3.11315514e-02  0.00000000e+00
  -6.41677468e-03  2.29271606e-03  3.38308568e-03 -6.46594197e-03]]
intercept: [-0.36083029]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_3.pickle')
test_set = test_set.drop(columns=[*_cols_bin, *_cols_real])
test_set.shape

(153993, 753)

# test classifier
predic = best_model.predict(test_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
print("predictions:\n", predic)

predictions:
 [1 0 1 ... 0 0 0]

set(predic)

{0, 1}

# evaluate classifier

print("Report for Support Vector Machine:")
print(_classification_report(test_set['likes'], predic))

print("Accuracy for Support Vector Machine:", _accuracy_score(test_set['likes'], predic)*100)

Report for Support Vector Machine:
              precision    recall  f1-score   support

           0       0.70      0.36      0.48     50930
           1       0.75      0.92      0.83    103063

    accuracy                           0.74    153993
   macro avg       0.72      0.64      0.65    153993
weighted avg       0.73      0.74      0.71    153993

Accuracy for Support Vector Machine: 73.79556213594124

# Confusion matrix for SVC

print("Confusion Matrix for SVC: ")
_confusion_matrix(test_set['likes'], predic)

Confusion Matrix for SVC:

array([[18342, 32588],
       [ 7765, 95298]], dtype=int64)

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], predic)

_plt.plot(fpr,tpr)
_plt.xlim([0.0,1.0])
_plt.ylim([0.0,1.0])

_plt.title("SVM ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_del_all()

7.8 Random forest only standard columns¶

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')
train_set.shape

(558386, 787)

train_set = train_set.drop(columns=[*_cols_bin, *_cols_real])
train_set.shape

(558386, 753)

params = _jl.load("../models/best_Random_Forest_2.joblib").get_params()
params['n_jobs'] = -1
params['verbose'] = 10
best_model = _RandomForestClassifier(**params)
best_model.get_params()

{'bootstrap': False,
 'class_weight': None,
 'criterion': 'entropy',
 'max_depth': 50,
 'max_features': 'auto',
 'max_leaf_nodes': None,
 'min_impurity_decrease': 0.0,
 'min_impurity_split': None,
 'min_samples_leaf': 2,
 'min_samples_split': 10,
 'min_weight_fraction_leaf': 0.0,
 'n_estimators': 1000,
 'n_jobs': -1,
 'oob_score': False,
 'random_state': None,
 'verbose': 10,
 'warm_start': False}

best_model.fit(train_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])

[Parallel(n_jobs=-1)]: Using backend ThreadingBackend with 12 concurrent workers.

building tree 1 of 1000building tree 2 of 1000
building tree 3 of 1000

building tree 4 of 1000
building tree 5 of 1000
building tree 6 of 1000
building tree 7 of 1000
building tree 8 of 1000
building tree 9 of 1000
building tree 10 of 1000
building tree 11 of 1000
building tree 12 of 1000
building tree 13 of 1000

[Parallel(n_jobs=-1)]: Done   1 tasks      | elapsed:   20.9s

building tree 14 of 1000
building tree 15 of 1000
building tree 16 of 1000
building tree 17 of 1000
building tree 18 of 1000
building tree 19 of 1000
building tree 20 of 1000

[Parallel(n_jobs=-1)]: Done   8 tasks      | elapsed:   23.6s

building tree 21 of 1000
building tree 22 of 1000
building tree 23 of 1000
building tree 24 of 1000
building tree 25 of 1000
building tree 26 of 1000
building tree 27 of 1000
building tree 28 of 1000
building tree 29 of 1000

[Parallel(n_jobs=-1)]: Done  17 tasks      | elapsed:   44.0s

building tree 30 of 1000
building tree 31 of 1000
building tree 32 of 1000
building tree 33 of 1000
building tree 34 of 1000
building tree 35 of 1000
building tree 36 of 1000
building tree 37 of 1000
building tree 38 of 1000

[Parallel(n_jobs=-1)]: Done  26 tasks      | elapsed:  1.1min

building tree 39 of 1000
building tree 40 of 1000
building tree 41 of 1000
building tree 42 of 1000
building tree 43 of 1000
building tree 44 of 1000
building tree 45 of 1000
building tree 46 of 1000
building tree 47 of 1000
building tree 48 of 1000

_jl.dump(best_model, "../models/best_Random_Forest_std.joblib")

['../models/best_Random_Forest_std.joblib']

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_3.pickle')
test_set = test_set.drop(columns=[*_cols_bin, *_cols_real])
test_set.shape

(153993, 753)

# test classifier
predic = best_model.predict(test_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
print("predictions:\n", predic)

[Parallel(n_jobs=12)]: Using backend ThreadingBackend with 12 concurrent workers.
[Parallel(n_jobs=12)]: Done   1 tasks      | elapsed:    0.1s
[Parallel(n_jobs=12)]: Done   8 tasks      | elapsed:    0.2s
[Parallel(n_jobs=12)]: Done  17 tasks      | elapsed:    0.4s
[Parallel(n_jobs=12)]: Done  26 tasks      | elapsed:    0.6s
[Parallel(n_jobs=12)]: Done  37 tasks      | elapsed:    0.9s
[Parallel(n_jobs=12)]: Done  48 tasks      | elapsed:    1.0s
[Parallel(n_jobs=12)]: Done  61 tasks      | elapsed:    1.5s
[Parallel(n_jobs=12)]: Done  74 tasks      | elapsed:    1.7s
[Parallel(n_jobs=12)]: Done  89 tasks      | elapsed:    2.0s
[Parallel(n_jobs=12)]: Done 104 tasks      | elapsed:    2.4s
[Parallel(n_jobs=12)]: Done 121 tasks      | elapsed:    2.8s
[Parallel(n_jobs=12)]: Done 138 tasks      | elapsed:    3.2s
[Parallel(n_jobs=12)]: Done 157 tasks      | elapsed:    3.6s
[Parallel(n_jobs=12)]: Done 176 tasks      | elapsed:    4.1s
[Parallel(n_jobs=12)]: Done 197 tasks      | elapsed:    4.6s
[Parallel(n_jobs=12)]: Done 218 tasks      | elapsed:    5.1s
[Parallel(n_jobs=12)]: Done 241 tasks      | elapsed:    5.7s
[Parallel(n_jobs=12)]: Done 264 tasks      | elapsed:    6.2s
[Parallel(n_jobs=12)]: Done 289 tasks      | elapsed:    6.8s
[Parallel(n_jobs=12)]: Done 314 tasks      | elapsed:    7.4s
[Parallel(n_jobs=12)]: Done 341 tasks      | elapsed:    8.0s
[Parallel(n_jobs=12)]: Done 368 tasks      | elapsed:    8.6s
[Parallel(n_jobs=12)]: Done 397 tasks      | elapsed:    9.2s
[Parallel(n_jobs=12)]: Done 426 tasks      | elapsed:   10.0s
[Parallel(n_jobs=12)]: Done 457 tasks      | elapsed:   10.6s
[Parallel(n_jobs=12)]: Done 488 tasks      | elapsed:   11.4s
[Parallel(n_jobs=12)]: Done 521 tasks      | elapsed:   12.2s
[Parallel(n_jobs=12)]: Done 554 tasks      | elapsed:   13.0s
[Parallel(n_jobs=12)]: Done 589 tasks      | elapsed:   13.8s
[Parallel(n_jobs=12)]: Done 624 tasks      | elapsed:   14.7s
[Parallel(n_jobs=12)]: Done 661 tasks      | elapsed:   15.6s
[Parallel(n_jobs=12)]: Done 698 tasks      | elapsed:   16.5s
[Parallel(n_jobs=12)]: Done 737 tasks      | elapsed:   17.4s
[Parallel(n_jobs=12)]: Done 776 tasks      | elapsed:   18.3s
[Parallel(n_jobs=12)]: Done 817 tasks      | elapsed:   19.3s
[Parallel(n_jobs=12)]: Done 858 tasks      | elapsed:   20.2s
[Parallel(n_jobs=12)]: Done 901 tasks      | elapsed:   21.2s
[Parallel(n_jobs=12)]: Done 944 tasks      | elapsed:   22.2s

predictions:
 [0 0 1 ... 1 1 0]

[Parallel(n_jobs=12)]: Done 1000 out of 1000 | elapsed:   23.5s finished

set(predic)

{0, 1}

# evaluate classifier

print("Report for Random Forest classifier:")
print(_classification_report(test_set['likes'], predic))

print("Accuracy for Random Forest Classifier:", _accuracy_score(test_set['likes'], predic)*100)

Report for Random Forest classifier:
              precision    recall  f1-score   support

           0       0.69      0.40      0.51     50930
           1       0.75      0.91      0.83    103063

    accuracy                           0.74    153993
   macro avg       0.72      0.66      0.67    153993
weighted avg       0.73      0.74      0.72    153993

Accuracy for Random Forest Classifier: 74.30337742624666

# Confusion matrix for Random Forest

print("Confusion Matrix for Random Forest: ")
_confusion_matrix(test_set['likes'], predic)

Confusion Matrix for Random Forest:

array([[20322, 30608],
       [ 8963, 94100]], dtype=int64)

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], predic)

_plt.plot(fpr,tpr)
_plt.xlim([0.0,1.0])
_plt.ylim([0.0,1.0])

_plt.title("Random Forest ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_del_all()

7.9 Neural Network only standard columns¶

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')
train_set.shape

(558386, 787)

train_set = train_set.drop(columns=[*_cols_bin, *_cols_real])
train_set.shape

(558386, 753)

# Number of neurons for hiddens layer
number_samples, number_features = train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']).shape

number_input_neurons = number_features
number_output_neurons = 1
number_train_semples = number_samples
alpha = 7

number_hidden_neurons = round(number_train_semples / (alpha * (number_input_neurons + number_output_neurons)))
print(number_hidden_neurons)

107

classifier = _Sequential()

# First Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal', input_dim = number_features))

# Second Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Third Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fourth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fifth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Output Layer
classifier.add(_Dense(1, activation = 'sigmoid', kernel_initializer = 'random_normal'))

# Compiling the neural network
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

# Fitting the data to the training dataset
classifier.fit(train_set.drop(
    columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']),
    train_set['likes'], validation_split = 0.3, batch_size = 100, epochs = 100)

Train on 390870 samples, validate on 167516 samples
Epoch 1/100
390870/390870 [==============================] - 34s 86us/step - loss: 0.5951 - acc: 0.6957 - val_loss: 0.5575 - val_acc: 0.7210
Epoch 2/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5558 - acc: 0.7236 - val_loss: 0.5577 - val_acc: 0.7252
Epoch 3/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5467 - acc: 0.7305 - val_loss: 0.5468 - val_acc: 0.7331
Epoch 4/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5421 - acc: 0.7342 - val_loss: 0.5335 - val_acc: 0.7436
Epoch 5/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5401 - acc: 0.7360 - val_loss: 0.5426 - val_acc: 0.7396
Epoch 6/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5375 - acc: 0.7379 - val_loss: 0.5408 - val_acc: 0.7385
Epoch 7/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5345 - acc: 0.7396 - val_loss: 0.5380 - val_acc: 0.7406
Epoch 8/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5344 - acc: 0.7400 - val_loss: 0.5378 - val_acc: 0.7397
Epoch 9/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5331 - acc: 0.7408 - val_loss: 0.5368 - val_acc: 0.7408
Epoch 10/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5311 - acc: 0.7417 - val_loss: 0.5317 - val_acc: 0.7421
Epoch 11/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5297 - acc: 0.7425 - val_loss: 0.5350 - val_acc: 0.7427
Epoch 12/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5293 - acc: 0.7432 - val_loss: 0.5306 - val_acc: 0.7439
Epoch 13/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5284 - acc: 0.7442 - val_loss: 0.5353 - val_acc: 0.7400
Epoch 14/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5264 - acc: 0.7447 - val_loss: 0.5302 - val_acc: 0.7443
Epoch 15/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5266 - acc: 0.7448 - val_loss: 0.5313 - val_acc: 0.7454
Epoch 16/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5255 - acc: 0.7454 - val_loss: 0.5265 - val_acc: 0.7478
Epoch 17/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5247 - acc: 0.7457 - val_loss: 0.5323 - val_acc: 0.7428
Epoch 18/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5243 - acc: 0.7459 - val_loss: 0.5336 - val_acc: 0.7459
Epoch 19/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5238 - acc: 0.7465 - val_loss: 0.5264 - val_acc: 0.7475
Epoch 20/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5227 - acc: 0.7468 - val_loss: 0.5310 - val_acc: 0.7445
Epoch 21/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5230 - acc: 0.7469 - val_loss: 0.5320 - val_acc: 0.7427
Epoch 22/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5229 - acc: 0.7466 - val_loss: 0.5262 - val_acc: 0.7453
Epoch 23/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5214 - acc: 0.7479 - val_loss: 0.5291 - val_acc: 0.7446
Epoch 24/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5208 - acc: 0.7480 - val_loss: 0.5219 - val_acc: 0.7507
Epoch 25/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5203 - acc: 0.7487 - val_loss: 0.5378 - val_acc: 0.7360
Epoch 26/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5201 - acc: 0.7483 - val_loss: 0.5325 - val_acc: 0.7439
Epoch 27/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5196 - acc: 0.7490 - val_loss: 0.5261 - val_acc: 0.7464
Epoch 28/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5195 - acc: 0.7486 - val_loss: 0.5202 - val_acc: 0.7501
Epoch 29/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5198 - acc: 0.7488 - val_loss: 0.5267 - val_acc: 0.7473
Epoch 30/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5186 - acc: 0.7497 - val_loss: 0.5273 - val_acc: 0.7444
Epoch 31/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5187 - acc: 0.7495 - val_loss: 0.5261 - val_acc: 0.7458
Epoch 32/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5179 - acc: 0.7494 - val_loss: 0.5359 - val_acc: 0.7392
Epoch 33/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5193 - acc: 0.7497 - val_loss: 0.5280 - val_acc: 0.7448
Epoch 34/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5172 - acc: 0.7500 - val_loss: 0.5214 - val_acc: 0.7509
Epoch 35/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5173 - acc: 0.7497 - val_loss: 0.5247 - val_acc: 0.7478
Epoch 36/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5171 - acc: 0.7500 - val_loss: 0.5184 - val_acc: 0.7517
Epoch 37/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5174 - acc: 0.7498 - val_loss: 0.5223 - val_acc: 0.7500
Epoch 38/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5170 - acc: 0.7498 - val_loss: 0.5207 - val_acc: 0.7491
Epoch 39/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5165 - acc: 0.7502 - val_loss: 0.5375 - val_acc: 0.7415
Epoch 40/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5156 - acc: 0.7508 - val_loss: 0.5212 - val_acc: 0.7486
Epoch 41/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5153 - acc: 0.7510 - val_loss: 0.5206 - val_acc: 0.7479
Epoch 42/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5151 - acc: 0.7517 - val_loss: 0.5196 - val_acc: 0.7513
Epoch 43/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5152 - acc: 0.7513 - val_loss: 0.5193 - val_acc: 0.7509
Epoch 44/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5149 - acc: 0.7517 - val_loss: 0.5192 - val_acc: 0.7515
Epoch 45/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5146 - acc: 0.7518 - val_loss: 0.5285 - val_acc: 0.7459
Epoch 46/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5149 - acc: 0.7514 - val_loss: 0.5183 - val_acc: 0.7517
Epoch 47/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5148 - acc: 0.7514 - val_loss: 0.5197 - val_acc: 0.7514
Epoch 48/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5144 - acc: 0.7516 - val_loss: 0.5254 - val_acc: 0.7486
Epoch 49/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5141 - acc: 0.7517 - val_loss: 0.5229 - val_acc: 0.7511
Epoch 50/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5134 - acc: 0.7523 - val_loss: 0.5231 - val_acc: 0.7497
Epoch 51/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5128 - acc: 0.7528 - val_loss: 0.5192 - val_acc: 0.7512
Epoch 52/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5132 - acc: 0.7522 - val_loss: 0.5211 - val_acc: 0.7492
Epoch 53/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5159 - acc: 0.7510 - val_loss: 0.5260 - val_acc: 0.7476
Epoch 54/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5197 - acc: 0.7466 - val_loss: 0.5467 - val_acc: 0.7242
Epoch 55/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5188 - acc: 0.7477 - val_loss: 0.5220 - val_acc: 0.7509
Epoch 56/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5135 - acc: 0.7520 - val_loss: 0.5221 - val_acc: 0.7520
Epoch 57/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5125 - acc: 0.7535 - val_loss: 0.5214 - val_acc: 0.7504
Epoch 58/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5123 - acc: 0.7530 - val_loss: 0.5221 - val_acc: 0.7510
Epoch 59/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5116 - acc: 0.7529 - val_loss: 0.5278 - val_acc: 0.7474
Epoch 60/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5116 - acc: 0.7530 - val_loss: 0.5270 - val_acc: 0.7475
Epoch 61/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5112 - acc: 0.7533 - val_loss: 0.5216 - val_acc: 0.7497
Epoch 62/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5111 - acc: 0.7535 - val_loss: 0.5195 - val_acc: 0.7503
Epoch 63/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5112 - acc: 0.7531 - val_loss: 0.5194 - val_acc: 0.7516
Epoch 64/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5108 - acc: 0.7534 - val_loss: 0.5210 - val_acc: 0.7499
Epoch 65/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5109 - acc: 0.7537 - val_loss: 0.5206 - val_acc: 0.7512
Epoch 66/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5127 - acc: 0.7526 - val_loss: 0.5201 - val_acc: 0.7504
Epoch 67/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5106 - acc: 0.7537 - val_loss: 0.5275 - val_acc: 0.7466
Epoch 68/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5103 - acc: 0.7539 - val_loss: 0.5198 - val_acc: 0.7493
Epoch 69/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5118 - acc: 0.7537 - val_loss: 0.5189 - val_acc: 0.7532
Epoch 70/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5104 - acc: 0.7542 - val_loss: 0.5242 - val_acc: 0.7473
Epoch 71/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5100 - acc: 0.7541 - val_loss: 0.5208 - val_acc: 0.7516
Epoch 72/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5100 - acc: 0.7536 - val_loss: 0.5187 - val_acc: 0.7512
Epoch 73/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5107 - acc: 0.7538 - val_loss: 0.5194 - val_acc: 0.7506
Epoch 74/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5095 - acc: 0.7543 - val_loss: 0.5203 - val_acc: 0.7520
Epoch 75/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5096 - acc: 0.7543 - val_loss: 0.5243 - val_acc: 0.7492
Epoch 76/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5105 - acc: 0.7543 - val_loss: 0.5231 - val_acc: 0.7486
Epoch 77/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5092 - acc: 0.7544 - val_loss: 0.5211 - val_acc: 0.7492
Epoch 78/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5092 - acc: 0.7545 - val_loss: 0.5191 - val_acc: 0.7538
Epoch 79/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5087 - acc: 0.7550 - val_loss: 0.5187 - val_acc: 0.7520
Epoch 80/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5088 - acc: 0.7550 - val_loss: 0.5203 - val_acc: 0.7517
Epoch 81/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5090 - acc: 0.7547 - val_loss: 0.5209 - val_acc: 0.7491
Epoch 82/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5084 - acc: 0.7551 - val_loss: 0.5226 - val_acc: 0.7492
Epoch 83/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5095 - acc: 0.7547 - val_loss: 0.5281 - val_acc: 0.7500
Epoch 84/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5089 - acc: 0.7548 - val_loss: 0.5309 - val_acc: 0.7432
Epoch 85/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5100 - acc: 0.7535 - val_loss: 0.5200 - val_acc: 0.7509
Epoch 86/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5088 - acc: 0.7552 - val_loss: 0.5185 - val_acc: 0.7534
Epoch 87/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5087 - acc: 0.7549 - val_loss: 0.5182 - val_acc: 0.7525
Epoch 88/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5081 - acc: 0.7551 - val_loss: 0.5224 - val_acc: 0.7507
Epoch 89/100
390870/390870 [==============================] - 25s 64us/step - loss: 0.5091 - acc: 0.7548 - val_loss: 0.5283 - val_acc: 0.7477
Epoch 90/100
390870/390870 [==============================] - 25s 65us/step - loss: 0.5080 - acc: 0.7553 - val_loss: 0.5197 - val_acc: 0.7509
Epoch 91/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5077 - acc: 0.7556 - val_loss: 0.5179 - val_acc: 0.7522
Epoch 92/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5075 - acc: 0.7556 - val_loss: 0.5288 - val_acc: 0.7469
Epoch 93/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5087 - acc: 0.7553 - val_loss: 0.5216 - val_acc: 0.7517
Epoch 94/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5096 - acc: 0.7551 - val_loss: 0.5242 - val_acc: 0.7483
Epoch 95/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5096 - acc: 0.7551 - val_loss: 0.5217 - val_acc: 0.7505
Epoch 96/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5102 - acc: 0.7536 - val_loss: 0.5292 - val_acc: 0.7473
Epoch 97/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5079 - acc: 0.7557 - val_loss: 0.5224 - val_acc: 0.7502
Epoch 98/100
390870/390870 [==============================] - 26s 65us/step - loss: 0.5079 - acc: 0.7552 - val_loss: 0.5220 - val_acc: 0.7524
Epoch 99/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5069 - acc: 0.7559 - val_loss: 0.5239 - val_acc: 0.7496
Epoch 100/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5140 - acc: 0.7515 - val_loss: 0.5198 - val_acc: 0.7499

<keras.callbacks.History at 0x1b708c73eb8>

# Save trained model
classifier.save("../models/trained_deep_neural_network_std.h5")

# evaluation of the model training
evaluation = classifier.evaluate(train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])
print(evaluation)

558386/558386 [==============================] - 17s 31us/step
[0.5116986021793278, 0.7529630040865652]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_3.pickle')
test_set = test_set.drop(columns=[*_cols_bin, *_cols_real])
test_set.shape

(153993, 753)

prediction = classifier.predict(test_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
binary_prediction = _binarize(prediction, threshold = 0.5)
binary_prediction

array([[1.],
       [0.],
       [0.],
       ...,
       [0.],
       [1.],
       [0.]], dtype=float32)

set(binary_prediction[:,0])

{0.0, 1.0}

print("Report for Deep Learning approach:")
print(_classification_report(test_set['likes'], binary_prediction))

print("Accuracy for Deep Learning approach:", _accuracy_score(test_set['likes'], binary_prediction) * 100)

Report for Deep Learning approach:
              precision    recall  f1-score   support

           0       0.72      0.36      0.48     50930
           1       0.75      0.93      0.83    103063

    accuracy                           0.74    153993
   macro avg       0.73      0.64      0.65    153993
weighted avg       0.74      0.74      0.71    153993

Accuracy for Deep Learning approach: 74.13713610358913

matrix = _confusion_matrix(test_set['likes'], binary_prediction)
print(matrix)

[[18245 32685]
 [ 7142 95921]]

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], binary_prediction)

_plt.plot(fpr, tpr)
_plt.xlim([0.0, 1.0])
_plt.ylim([0.0, 1.0])

_plt.title("Neural Network ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_del_all()

7.10 SVM only real columns¶

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')
train_set.shape

(558386, 787)

train_set = train_set.drop(columns=[*_cols_bin, *_cols_std])
train_set.shape

(558386, 754)

best_model = _jl.load("../models/best_SVM.joblib")
best_model.set_params(verbose=10)
best_model.get_params()

{'C': 0.001,
 'class_weight': None,
 'dual': True,
 'fit_intercept': True,
 'intercept_scaling': 1,
 'loss': 'squared_hinge',
 'max_iter': 50000,
 'multi_class': 'ovr',
 'penalty': 'l2',
 'random_state': 0,
 'tol': 0.0001,
 'verbose': 10}

best_model.fit(train_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])

[LibLinear]

C:\Users\super\Anaconda3\envs\tf-gpu\lib\site-packages\sklearn\svm\base.py:929: ConvergenceWarning: Liblinear failed to converge, increase the number of iterations.
  "the number of iterations.", ConvergenceWarning)

LinearSVC(C=0.001, class_weight=None, dual=True, fit_intercept=True,
          intercept_scaling=1, loss='squared_hinge', max_iter=50000,
          multi_class='ovr', penalty='l2', random_state=0, tol=0.0001,
          verbose=10)

_jl.dump(best_model, "../models/best_SVM_real.joblib")

['../models/best_SVM_real.joblib']

print("coef:", best_model.coef_)
print("intercept:", best_model.intercept_)

coef: [[-1.43716297e-01 -2.04535344e-01  3.42412756e-01 -1.70106570e-01
   9.42906076e-02 -7.64244164e-02  2.83228461e-05 -3.44367883e-02
   7.43154614e-03 -2.10748750e-05  6.42141961e-01 -7.67363295e-05
   1.05147224e-03  2.61865749e-04 -7.96263673e-04 -5.67863277e-04
  -6.66914030e-04  1.09450014e-03 -1.87557776e-04 -8.91209472e-05
   4.19307492e-04 -4.27875482e-05  8.48324927e-05 -8.68450895e-06
  -2.98160509e-04 -1.31943690e-01  4.98907004e-05 -4.09473030e-02
   2.69595801e-04  1.66887312e-02 -4.39124340e-04 -2.50972181e-04
   4.38847068e-05  1.61532968e-04 -9.42543009e-05  6.55236412e-02
  -6.76947062e-04  3.02737339e-02  2.37219033e-02  3.53476493e-02
   4.01342657e-02  5.32870518e-02  4.85615621e-02  1.44103811e-02
   2.56643390e-02  2.03262701e-02  4.17940324e-02 -1.24827229e-01
  -9.21161254e-02 -1.28367851e-01 -7.71161849e-02 -1.25495132e-01
  -1.42699889e-01 -1.26632389e-01 -1.07560837e-01 -1.11117980e-01
  -1.19877143e-01 -1.14195865e-01 -1.11238198e-01 -8.52179200e-02
  -8.51377622e-02 -9.34756944e-02 -8.14798291e-02 -8.91023815e-02
  -9.11799522e-02 -8.48363095e-02 -8.01925625e-02  6.83784341e-03
   1.92254971e-02  1.40058851e-02  2.39106028e-02  2.46003929e-02
  -2.27673372e-02  1.29762647e-02  3.11092757e-02 -5.74496098e-02
   4.47713432e-02  2.02799171e-02 -7.57741485e-02 -1.44389973e-02
   3.17057017e-02  2.86103761e-02  6.51142500e-02 -8.49139531e-03
  -7.77972563e-02 -1.30119107e-02 -3.40256582e-02 -1.44796203e-03
  -1.38423057e-03  1.31105830e-02  2.60418460e-02  2.33846729e-03
   5.29074929e-03  3.48008783e-01 -9.74783977e-03  1.77673098e-02
  -8.86489053e-03  1.80711141e-02  1.57005941e-02  4.03685529e-02
  -1.99766603e-02  2.17479883e-02 -1.81747456e-01  1.99719398e-02
   5.82290558e-02 -2.43132901e-02 -1.15618867e-01  8.72118852e-04
  -8.60105516e-02  2.41831332e-02 -2.10875723e-02 -5.56974893e-02
   1.37147410e-02  7.75777631e-03  1.20550494e-02 -8.02849787e-03
   1.97158718e-02 -6.76034995e-02  1.76035986e-02  1.75296613e-02
   7.07208171e-03  3.34970610e-02 -1.20573798e-01  8.31427170e-03
  -1.02979836e-01 -2.38829799e-02  1.64392594e-03 -7.94164576e-03
  -1.32036699e-02  6.66926944e-02  2.11093298e-02  2.92684975e-02
   1.99726243e-02 -9.68394996e-02  2.19072433e-02 -3.34938244e-02
  -7.81002948e-02  1.36871377e-02 -4.44362689e-02  3.37312734e-02
  -1.78446481e-02  6.16230824e-02  9.12907609e-02  1.73681006e-02
   9.42710477e-03  1.21382260e-02 -1.07187657e-02  3.71220243e-03
  -6.58497330e-03  1.37964168e-02  1.29674903e-01 -1.74803903e-02
  -1.57516188e-02 -1.36993620e-01 -9.85035233e-03 -2.89222187e-02
   7.58170440e-03  4.36057261e-02  6.97113801e-02 -2.81749802e-02
   2.09981262e-02  1.16701922e-01 -2.16218691e-02 -3.76903594e-02
   2.89249867e-02 -2.17925185e-02 -3.30512837e-02 -3.79864486e-03
   5.27191509e-02 -1.21867205e-02 -2.28182341e-02 -8.33960295e-02
   1.41297582e-02 -2.23504605e-02 -1.32998285e-02  7.90823855e-02
  -1.83089317e-02  2.10069237e-02  4.21856531e-03 -8.23935593e-02
   3.15873915e-02 -1.00053202e-02 -6.13077348e-02  1.35207139e-02
  -7.91516814e-03 -5.59031582e-03 -5.71692743e-03 -3.72606011e-03
  -2.37534039e-02  6.98443308e-02 -1.74441860e-02 -8.53985002e-03
   1.71990647e-02 -1.10963279e-01 -1.13059523e-02  1.18665664e-02
   4.83008993e-03 -1.30285498e-02 -8.77128617e-02  9.83830543e-03
   1.84587386e-02 -4.81211451e-02 -1.50782933e-02 -2.94999297e-04
  -2.95794368e-02  4.77008288e-02 -1.60578875e-02  4.29225749e-02
   1.56618451e-02 -6.85062777e-03  3.59512189e-02  3.35828023e-02
  -8.54411962e-03 -1.37198962e-02  4.75972657e-02  1.17362096e-02
  -5.43824915e-02 -3.19341547e-02 -3.71196861e-02 -3.30255707e-02
  -8.67861270e-02  6.77269315e-03 -6.64047346e-03 -3.21719005e-03
  -1.98840205e-03  2.39095825e-02  5.76495663e-03 -4.35348076e-02
  -1.18309881e-02  6.97192193e-02  1.69856151e-02  2.50381925e-02
  -5.02032740e-02 -4.71014503e-02  7.79945547e-03 -2.66495190e-02
   5.29165693e-03  3.96304236e-03  8.36005323e-03 -1.80505866e-02
   2.92364047e-02 -8.79619351e-03 -5.24025156e-02 -8.38200860e-03
   4.44280946e-03  9.30483463e-03 -3.83530258e-02 -3.02118791e-02
  -1.54155851e-02 -5.90653519e-04 -5.12051389e-02  8.42693854e-03
   1.65402105e-02  1.57442789e-02  1.31571247e-02  2.07821529e-02
  -1.17494214e-03 -5.28376198e-03 -1.99382278e-02 -2.47230047e-03
  -5.39078174e-03  5.40381038e-03 -6.14568668e-03 -4.77322227e-02
  -8.52581213e-04 -3.17905886e-03  6.64875505e-03 -1.07021403e-02
  -6.19146130e-02 -6.37739266e-03  4.45299066e-02 -1.15002674e-02
   4.38282778e-03 -6.36030533e-02  1.20311051e-02  2.59641550e-03
   3.86870102e-03  2.42142186e-02  4.84045567e-03 -5.89914590e-03
  -2.75915844e-02 -5.10269072e-05  3.28629430e-04  1.33020475e-02
   1.68775734e-02 -1.12454827e-02 -4.82852523e-02 -4.44387073e-04
   1.95298920e-02  4.51779966e-03  2.45641205e-02 -4.67522475e-03
   1.43765356e-02 -2.70234675e-03 -5.00163882e-02  6.22747213e-03
  -2.08606834e-02  3.05999934e-02 -3.98381028e-02 -2.47981151e-02
  -7.95026123e-05  1.09004063e-02  2.09445287e-02 -2.12375146e-02
  -2.31301644e-03 -9.02606034e-02 -9.50105319e-02  5.42986944e-03
   4.44964730e-03 -4.18983184e-03  2.81625116e-02 -1.25999766e-02
  -2.41662714e-03  2.55381467e-03 -4.06825386e-02  6.15082758e-03
  -1.68552411e-02  3.95541394e-02  1.04655343e-02  4.83088682e-02
  -1.43646018e-02  1.85676225e-02  4.80610771e-02  5.83198667e-03
   1.88030789e-02 -9.92599369e-03  1.92759739e-02  9.68732093e-03
   3.60602287e-03  2.11169130e-03  7.12049488e-03  5.77638547e-02
   1.48541561e-02  1.54975082e-03  2.00805555e-03  1.74872561e-02
  -2.20886530e-02  2.30314189e-02  1.32497217e-02 -5.49148428e-03
   3.53915515e-03  8.84630798e-03  3.94741276e-03 -2.96144230e-04
  -2.62217011e-02  1.46662615e-02  4.43793543e-03  8.37307760e-03
  -8.88604631e-03 -1.46383121e-02 -4.70885434e-02 -1.08056738e-02
   2.63550904e-02  4.25842720e-02 -1.06471586e-02 -3.20178407e-03
  -4.00402304e-02  7.74943513e-03  1.30340136e-02 -5.33402478e-03
   1.63923629e-02 -5.74304402e-04  4.87202627e-03  3.33034034e-02
   1.86518308e-02  8.05365248e-03  2.16511024e-02  1.48769671e-04
   2.51305937e-02 -1.58727100e-02  4.76463924e-03 -5.93707391e-02
  -8.57548415e-03 -2.15737399e-03 -1.51769019e-02 -1.06675964e-02
   9.10391321e-03  8.09320421e-03 -1.00892549e-02 -1.21250581e-02
   1.94932467e-02 -1.88602889e-02 -1.77840807e-02 -4.17600384e-02
   3.41119549e-04 -1.50481753e-03 -3.93696545e-03  8.32069882e-04
  -1.00086792e-02 -3.08954332e-02  1.50102907e-02 -2.02094891e-02
   1.58622884e-02 -3.27257295e-02 -2.90866697e-02  2.49579263e-02
  -2.65487658e-02  3.68630700e-02  4.06325061e-02 -1.15374121e-02
  -8.61179632e-03  3.32551687e-03 -4.52654593e-02 -1.36332587e-02
   1.22906007e-02 -1.88976446e-02  1.27437107e-02  6.49445940e-02
  -9.66161494e-03  2.15711752e-02 -3.45778686e-02  5.80762743e-04
  -2.13839479e-04  4.34493773e-03  9.90380184e-04  1.08717932e-02
   2.97467351e-02 -5.94337775e-03 -2.60234637e-02 -7.95258176e-03
  -6.48299816e-03  1.95546414e-02  1.67365683e-02 -1.14618509e-02
  -3.58714600e-03 -2.54777520e-03 -2.84852661e-02  1.49303319e-03
   4.95068791e-02  1.60731206e-02 -1.74493711e-02 -1.81267180e-02
   2.36541572e-02  1.06580281e-02  1.65311243e-02 -1.61455394e-03
   8.72829086e-03  3.86870102e-03 -2.03087109e-03 -2.67693393e-03
   5.60706191e-04  6.02260553e-03 -2.39887876e-02  2.34089987e-02
   5.32528475e-03 -4.45201832e-03  1.46990462e-02 -5.90900136e-02
  -1.53664685e-02  2.64979387e-02 -3.49239256e-02 -3.38094692e-02
   7.92953814e-03  7.47241074e-02 -6.75645710e-03  2.06018288e-02
   1.29008264e-02  2.63497889e-02  3.09334173e-02  7.84939100e-02
   1.01068985e-02  6.25077315e-03  1.20308144e-02  7.39519295e-03
   1.24135458e-03  7.70286332e-04 -5.43830545e-03 -6.38783462e-03
  -1.35901699e-02  2.67742151e-02 -2.40821850e-02  1.82810151e-02
   1.24718095e-02 -7.62080333e-03  4.03348586e-02 -1.71184613e-03
  -2.70636989e-02  1.19119881e-02 -1.64302434e-03  2.00077971e-02
  -1.40176110e-02 -8.05835669e-03  9.67086176e-03 -5.18628634e-03
   1.93308559e-02 -2.89383369e-02 -4.29723599e-02  5.37350245e-02
   3.16453339e-03 -2.64130760e-02  2.77979932e-02 -3.14515490e-02
   1.84017732e-03 -2.73363357e-02  2.61905678e-02 -2.08372057e-02
   2.83759060e-03  1.82699625e-02 -1.45091797e-02 -7.85236472e-05
   7.31865718e-03 -4.21628925e-03  2.07016491e-02 -3.85148502e-02
  -1.89099266e-02 -8.58946052e-03  1.67151061e-02  3.49526992e-02
   4.18637313e-03 -3.59994317e-02  2.45687128e-02 -3.81881299e-03
  -1.06850579e-02 -5.22586771e-02  4.26735946e-02 -2.35013806e-03
   2.33494510e-02  9.71528304e-03 -1.07773945e-02  2.52467523e-03
  -3.04210778e-02 -1.73296552e-03 -4.49178273e-02  2.49230145e-02
   3.92718681e-03  1.22880860e-02  2.34955451e-02  2.65367107e-02
   1.44125852e-02 -3.69221456e-04  1.82773207e-02 -1.17576018e-02
   2.52757927e-02 -4.46373382e-02 -3.57136625e-04 -2.60679490e-05
  -1.57908492e-02 -3.45646799e-02  2.78523427e-02  7.07094907e-03
   4.48861252e-03 -3.83012103e-02  1.55599056e-02  3.70129306e-03
   1.15512975e-02 -9.01585174e-03  5.39040951e-05  1.66301005e-03
   5.38590657e-03 -1.92638044e-02 -2.24117148e-03 -2.78016044e-04
  -3.90414502e-02 -3.12567861e-02 -8.96096477e-04 -2.78328104e-02
   1.06643622e-02 -2.27281534e-03  3.16658561e-02 -3.20813228e-02
   6.39288622e-03  0.00000000e+00 -1.45071139e-02  1.42715026e-03
  -7.97438314e-03  3.84988264e-03  2.68315418e-03 -2.65674288e-02
  -4.05858718e-03  6.05797036e-03  7.80432532e-03 -1.91687863e-02
   6.11712672e-03  9.93745103e-03 -9.50309878e-04 -4.17014796e-02
  -1.06940936e-02 -5.59727389e-03 -2.88390609e-03 -3.31251482e-02
   7.56702389e-03  2.21977742e-02  6.03846263e-03  2.27310460e-02
  -3.58590645e-02 -6.25894466e-03  3.03627124e-02 -1.31549344e-02
   2.86701270e-02 -4.38543972e-03  5.96440014e-03  3.65200851e-02
  -3.44983708e-02 -3.54270601e-04 -1.17295962e-02 -1.29913844e-02
  -4.58019303e-03 -2.70093413e-02 -1.02683551e-03 -2.12796323e-02
  -6.28906254e-03 -2.13508993e-02 -1.31002531e-02  8.53354900e-03
  -1.01119559e-03  2.53777572e-03 -2.73862488e-02  5.13986485e-04
   4.96337590e-02  2.01269116e-03 -1.23967219e-02  5.99338545e-03
  -7.39463771e-03 -5.15502493e-02 -5.69401986e-02  8.36397356e-02
   1.49669026e-02  1.92547717e-02 -2.40386153e-03 -1.55658920e-02
  -4.15788011e-03  3.33850113e-03 -3.07508400e-02  1.30070473e-02
  -1.13289582e-02  0.00000000e+00  3.45435182e-03 -3.95516509e-02
  -2.22982560e-03 -3.48309267e-02 -7.31128590e-03  5.96541863e-03
   3.61227183e-03 -8.78392391e-03  1.37834712e-03 -6.69092052e-03
  -3.15195770e-02 -1.38635405e-02  1.00959273e-02  2.76226658e-03
   2.03101528e-04 -2.18130586e-02 -2.55485614e-02 -1.04517004e-03
  -1.51418225e-02 -3.12647177e-02 -4.59101212e-03 -4.15483850e-03
  -6.87723051e-04  1.96908603e-03  6.93292563e-03 -6.31346357e-05
   1.53845338e-02  3.17507328e-03  5.29435121e-03 -3.44138044e-03
   4.45274632e-03 -3.24623830e-02 -6.33364079e-02 -4.90524973e-02
   3.21685419e-02  6.39206044e-03 -3.37718688e-04  9.44383014e-04
   9.85393602e-04 -2.24601861e-03  1.42482016e-02  3.59703203e-03
  -5.04924862e-03  1.10421679e-02  6.41647623e-03  1.01527416e-02
   9.67600197e-04 -1.01501912e-02 -1.23796737e-04  5.19361132e-03
   3.57315292e-02 -4.99324726e-03 -7.82866781e-03 -6.62093182e-03
   1.08253961e-02  9.45856053e-03  1.93665333e-02  5.37784913e-03
  -6.54653955e-03 -1.00714377e-02  2.64630398e-02 -5.85493233e-04
   2.56982708e-02  3.34737700e-02  1.35830052e-03  1.21957786e-02
  -1.46874351e-02 -2.31088823e-03  9.95648118e-03 -4.27291467e-03
   8.88768511e-03 -1.22108668e-03  1.05574328e-02 -2.87483606e-02
   1.85895012e-03  2.89736005e-04  1.30702131e-02 -2.03961361e-02
  -2.20166959e-02 -1.72379361e-02 -3.75650065e-02  1.55585526e-02
   2.79515661e-02  7.45882964e-03  2.13700843e-02  1.30877206e-02
   2.53230154e-02 -2.08090208e-02 -7.53651928e-03 -3.25048945e-03
   3.19500741e-02 -1.35874362e-03  2.12078402e-02 -1.43182828e-02
   9.31946228e-03  1.82984277e-03  1.18614817e-02 -1.77738308e-02
  -6.14562184e-03  3.37351295e-02 -8.87142891e-04  1.04445750e-02
  -1.90250165e-02 -2.29921638e-03  1.08701737e-02  3.93571522e-02
  -2.64140804e-02 -2.62594850e-02  9.20156567e-04  1.21844794e-03
   1.57740349e-02  5.68792667e-04  7.02158502e-03 -4.43659971e-02
  -1.70125699e-03  1.79322546e-02  2.69916694e-02  3.34130488e-02
   0.00000000e+00 -5.87417755e-03  2.45578056e-03  3.08521576e-03
  -6.98591714e-03]]
intercept: [-0.34531121]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_3.pickle')
test_set = test_set.drop(columns=[*_cols_bin, *_cols_std])
test_set.shape

(153993, 754)

# test classifier
predic = best_model.predict(test_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
print("predictions:\n", predic)

predictions:
 [0 0 0 ... 0 0 0]

set(predic)

{0, 1}

# evaluate classifier

print("Report for Support Vector Machine:")
print(_classification_report(test_set['likes'], predic))

print("Accuracy for Support Vector Machine:", _accuracy_score(test_set['likes'], predic)*100)

Report for Support Vector Machine:
              precision    recall  f1-score   support

           0       0.55      0.49      0.52     50930
           1       0.76      0.81      0.78    103063

    accuracy                           0.70    153993
   macro avg       0.66      0.65      0.65    153993
weighted avg       0.69      0.70      0.69    153993

Accuracy for Support Vector Machine: 70.01292266531595

# Confusion matrix for SVC

print("Confusion Matrix for SVC: ")
_confusion_matrix(test_set['likes'], predic)

Confusion Matrix for SVC:

array([[24734, 26196],
       [19982, 83081]], dtype=int64)

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], predic)

_plt.plot(fpr,tpr)
_plt.xlim([0.0,1.0])
_plt.ylim([0.0,1.0])

_plt.title("SVM ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_del_all()

7.11 Random forest only real columns¶

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')
train_set.shape

(558386, 787)

train_set = train_set.drop(columns=[*_cols_bin, *_cols_std])
train_set.shape

(558386, 754)

params = _jl.load("../models/best_Random_Forest_2.joblib").get_params()
params['n_jobs'] = -1
params['verbose'] = 10
best_model = _RandomForestClassifier(**params)
best_model.get_params()

{'bootstrap': False,
 'class_weight': None,
 'criterion': 'entropy',
 'max_depth': 50,
 'max_features': 'auto',
 'max_leaf_nodes': None,
 'min_impurity_decrease': 0.0,
 'min_impurity_split': None,
 'min_samples_leaf': 2,
 'min_samples_split': 10,
 'min_weight_fraction_leaf': 0.0,
 'n_estimators': 1000,
 'n_jobs': -1,
 'oob_score': False,
 'random_state': None,
 'verbose': 10,
 'warm_start': False}

best_model.fit(train_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])

[Parallel(n_jobs=-1)]: Using backend ThreadingBackend with 12 concurrent workers.

building tree 1 of 1000building tree 2 of 1000building tree 3 of 1000building tree 4 of 1000

building tree 5 of 1000
building tree 6 of 1000
building tree 7 of 1000
building tree 8 of 1000
building tree 9 of 1000
building tree 10 of 1000
building tree 11 of 1000
building tree 12 of 1000


building tree 13 of 1000

[Parallel(n_jobs=-1)]: Done   1 tasks      | elapsed:   21.4s

building tree 14 of 1000
building tree 15 of 1000
building tree 16 of 1000
building tree 17 of 1000
building tree 18 of 1000
building tree 19 of 1000
building tree 20 of 1000
building tree 21 of 1000

[Parallel(n_jobs=-1)]: Done   8 tasks      | elapsed:   24.0s

building tree 22 of 1000
building tree 23 of 1000
building tree 24 of 1000
building tree 25 of 1000
building tree 26 of 1000
building tree 27 of 1000
building tree 28 of 1000

[Parallel(n_jobs=-1)]: Done  17 tasks      | elapsed:   46.1s

building tree 29 of 1000
building tree 30 of 1000
building tree 31 of 1000
building tree 32 of 1000
building tree 33 of 1000
building tree 34 of 1000
building tree 35 of 1000
building tree 36 of 1000
building tree 37 of 1000

[Parallel(n_jobs=-1)]: Done  26 tasks      | elapsed:  1.1min

building tree 38 of 1000
building tree 39 of 1000
building tree 40 of 1000
building tree 41 of 1000
building tree 42 of 1000
building tree 43 of 1000
building tree 44 of 1000
building tree 45 of 1000
building tree 46 of 1000
building tree 47 of 1000
building tree 48 of 1000

_jl.dump(best_model, "../models/best_Random_Forest_real.joblib")

['../models/best_Random_Forest_real.joblib']

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_3.pickle')
test_set = test_set.drop(columns=[*_cols_bin, *_cols_std])
test_set.shape

(153993, 754)

# test classifier
predic = best_model.predict(test_set.drop(columns=['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
print("predictions:\n", predic)

[Parallel(n_jobs=12)]: Using backend ThreadingBackend with 12 concurrent workers.
[Parallel(n_jobs=12)]: Done   1 tasks      | elapsed:    0.2s
[Parallel(n_jobs=12)]: Done   8 tasks      | elapsed:    0.2s
[Parallel(n_jobs=12)]: Done  17 tasks      | elapsed:    0.6s
[Parallel(n_jobs=12)]: Done  26 tasks      | elapsed:    0.9s
[Parallel(n_jobs=12)]: Done  37 tasks      | elapsed:    1.1s
[Parallel(n_jobs=12)]: Done  48 tasks      | elapsed:    1.4s
[Parallel(n_jobs=12)]: Done  61 tasks      | elapsed:    1.8s
[Parallel(n_jobs=12)]: Done  74 tasks      | elapsed:    2.1s
[Parallel(n_jobs=12)]: Done  89 tasks      | elapsed:    2.4s
[Parallel(n_jobs=12)]: Done 104 tasks      | elapsed:    2.7s
[Parallel(n_jobs=12)]: Done 121 tasks      | elapsed:    3.2s
[Parallel(n_jobs=12)]: Done 138 tasks      | elapsed:    3.6s
[Parallel(n_jobs=12)]: Done 157 tasks      | elapsed:    4.1s
[Parallel(n_jobs=12)]: Done 176 tasks      | elapsed:    4.5s
[Parallel(n_jobs=12)]: Done 197 tasks      | elapsed:    5.1s
[Parallel(n_jobs=12)]: Done 218 tasks      | elapsed:    5.6s
[Parallel(n_jobs=12)]: Done 241 tasks      | elapsed:    6.1s
[Parallel(n_jobs=12)]: Done 264 tasks      | elapsed:    6.7s
[Parallel(n_jobs=12)]: Done 289 tasks      | elapsed:    7.3s
[Parallel(n_jobs=12)]: Done 314 tasks      | elapsed:    7.9s
[Parallel(n_jobs=12)]: Done 341 tasks      | elapsed:    8.7s
[Parallel(n_jobs=12)]: Done 368 tasks      | elapsed:    9.3s
[Parallel(n_jobs=12)]: Done 397 tasks      | elapsed:   10.0s
[Parallel(n_jobs=12)]: Done 426 tasks      | elapsed:   10.7s
[Parallel(n_jobs=12)]: Done 457 tasks      | elapsed:   11.5s
[Parallel(n_jobs=12)]: Done 488 tasks      | elapsed:   12.2s
[Parallel(n_jobs=12)]: Done 521 tasks      | elapsed:   13.0s
[Parallel(n_jobs=12)]: Done 554 tasks      | elapsed:   13.8s
[Parallel(n_jobs=12)]: Done 589 tasks      | elapsed:   14.6s
[Parallel(n_jobs=12)]: Done 624 tasks      | elapsed:   15.5s
[Parallel(n_jobs=12)]: Done 661 tasks      | elapsed:   16.4s
[Parallel(n_jobs=12)]: Done 698 tasks      | elapsed:   17.2s
[Parallel(n_jobs=12)]: Done 737 tasks      | elapsed:   18.1s
[Parallel(n_jobs=12)]: Done 776 tasks      | elapsed:   19.1s
[Parallel(n_jobs=12)]: Done 817 tasks      | elapsed:   20.0s
[Parallel(n_jobs=12)]: Done 858 tasks      | elapsed:   21.0s
[Parallel(n_jobs=12)]: Done 901 tasks      | elapsed:   22.1s
[Parallel(n_jobs=12)]: Done 944 tasks      | elapsed:   23.0s
[Parallel(n_jobs=12)]: Done 1000 out of 1000 | elapsed:   24.3s finished

predictions:
 [0 0 1 ... 0 1 0]

set(predic)

{0, 1}

# evaluate classifier

print("Report for Random Forest classifier:")
print(_classification_report(test_set['likes'], predic))

print("Accuracy for Random Forest Classifier:", _accuracy_score(test_set['likes'], predic)*100)

Report for Random Forest classifier:
              precision    recall  f1-score   support

           0       0.70      0.39      0.50     50930
           1       0.75      0.92      0.83    103063

    accuracy                           0.74    153993
   macro avg       0.73      0.65      0.67    153993
weighted avg       0.74      0.74      0.72    153993

Accuracy for Random Forest Classifier: 74.34039209574462

# Confusion matrix for Random Forest

print("Confusion Matrix for Random Forest: ")
_confusion_matrix(test_set['likes'], predic)

Confusion Matrix for Random Forest:

array([[20055, 30875],
       [ 8639, 94424]], dtype=int64)

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], predic)

_plt.plot(fpr,tpr)
_plt.xlim([0.0,1.0])
_plt.ylim([0.0,1.0])

_plt.title("Random Forest ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_del_all()

7.12 Neural Network only real columns¶

train_set = _pd.read_pickle('../dataset/m2_n9/model_train_set_3.pickle')
train_set.shape

(558386, 787)

train_set = train_set.drop(columns=[*_cols_bin, *_cols_std])
train_set.shape

(558386, 754)

# Number of neurons for hiddens layer
number_samples, number_features = train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']).shape

number_input_neurons = number_features
number_output_neurons = 1
number_train_semples = number_samples
alpha = 7

number_hidden_neurons = round(number_train_semples / (alpha * (number_input_neurons + number_output_neurons)))
print(number_hidden_neurons)

106

classifier = _Sequential()

# First Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal', input_dim = number_features))

# Second Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Third Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fourth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Fifth Hidden Layer
classifier.add(_Dense(number_hidden_neurons, activation = 'relu', kernel_initializer = 'random_normal'))

# Output Layer
classifier.add(_Dense(1, activation = 'sigmoid', kernel_initializer = 'random_normal'))

# Compiling the neural network
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

# Fitting the data to the training dataset
classifier.fit(train_set.drop(
    columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']),
    train_set['likes'], validation_split = 0.3, batch_size = 100, epochs = 100)

Train on 390870 samples, validate on 167516 samples
Epoch 1/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5946 - acc: 0.6956 - val_loss: 0.5614 - val_acc: 0.7157
Epoch 2/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5553 - acc: 0.7237 - val_loss: 0.5431 - val_acc: 0.7351
Epoch 3/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5458 - acc: 0.7312 - val_loss: 0.5414 - val_acc: 0.7347
Epoch 4/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5414 - acc: 0.7351 - val_loss: 0.5421 - val_acc: 0.7341
Epoch 5/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5393 - acc: 0.7358 - val_loss: 0.5328 - val_acc: 0.7411
Epoch 6/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5361 - acc: 0.7378 - val_loss: 0.5395 - val_acc: 0.7355
Epoch 7/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5336 - acc: 0.7393 - val_loss: 0.5511 - val_acc: 0.7259
Epoch 8/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5327 - acc: 0.7404 - val_loss: 0.5319 - val_acc: 0.7417
Epoch 9/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5313 - acc: 0.7411 - val_loss: 0.5283 - val_acc: 0.7466
Epoch 10/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5300 - acc: 0.7423 - val_loss: 0.5301 - val_acc: 0.7469
Epoch 11/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5289 - acc: 0.7431 - val_loss: 0.5323 - val_acc: 0.7391
Epoch 12/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5282 - acc: 0.7433 - val_loss: 0.5353 - val_acc: 0.7420
Epoch 13/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5273 - acc: 0.7442 - val_loss: 0.5278 - val_acc: 0.7451
Epoch 14/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5261 - acc: 0.7449 - val_loss: 0.5282 - val_acc: 0.7424
Epoch 15/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5255 - acc: 0.7452 - val_loss: 0.5283 - val_acc: 0.7447
Epoch 16/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5248 - acc: 0.7460 - val_loss: 0.5224 - val_acc: 0.7485
Epoch 17/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5240 - acc: 0.7463 - val_loss: 0.5294 - val_acc: 0.7475
Epoch 18/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5233 - acc: 0.7464 - val_loss: 0.5351 - val_acc: 0.7403
Epoch 19/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5227 - acc: 0.7473 - val_loss: 0.5269 - val_acc: 0.7449
Epoch 20/100
390870/390870 [==============================] - 30s 76us/step - loss: 0.5221 - acc: 0.7478 - val_loss: 0.5241 - val_acc: 0.7486
Epoch 21/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5221 - acc: 0.7480 - val_loss: 0.5232 - val_acc: 0.7488
Epoch 22/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5209 - acc: 0.7480 - val_loss: 0.5202 - val_acc: 0.7505
Epoch 23/100
390870/390870 [==============================] - 26s 68us/step - loss: 0.5214 - acc: 0.7482 - val_loss: 0.5321 - val_acc: 0.7448
Epoch 24/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5197 - acc: 0.7491 - val_loss: 0.5218 - val_acc: 0.7495
Epoch 25/100
390870/390870 [==============================] - 30s 77us/step - loss: 0.5194 - acc: 0.7486 - val_loss: 0.5312 - val_acc: 0.7426
Epoch 26/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5186 - acc: 0.7492 - val_loss: 0.5195 - val_acc: 0.7495
Epoch 27/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5187 - acc: 0.7494 - val_loss: 0.5298 - val_acc: 0.7440
Epoch 28/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5180 - acc: 0.7498 - val_loss: 0.5346 - val_acc: 0.7420
Epoch 29/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5180 - acc: 0.7494 - val_loss: 0.5221 - val_acc: 0.7502
Epoch 30/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5176 - acc: 0.7495 - val_loss: 0.5265 - val_acc: 0.7498
Epoch 31/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5173 - acc: 0.7504 - val_loss: 0.5224 - val_acc: 0.7479
Epoch 32/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5174 - acc: 0.7503 - val_loss: 0.5295 - val_acc: 0.7468
Epoch 33/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5172 - acc: 0.7500 - val_loss: 0.5182 - val_acc: 0.7502
Epoch 34/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5171 - acc: 0.7502 - val_loss: 0.5211 - val_acc: 0.7487
Epoch 35/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5165 - acc: 0.7501 - val_loss: 0.5218 - val_acc: 0.7477
Epoch 36/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5168 - acc: 0.7505 - val_loss: 0.5328 - val_acc: 0.7411
Epoch 37/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5158 - acc: 0.7509 - val_loss: 0.5214 - val_acc: 0.7514
Epoch 38/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5151 - acc: 0.7518 - val_loss: 0.5209 - val_acc: 0.7505
Epoch 39/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5148 - acc: 0.7513 - val_loss: 0.5233 - val_acc: 0.7490
Epoch 40/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5147 - acc: 0.7516 - val_loss: 0.5204 - val_acc: 0.7511
Epoch 41/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5145 - acc: 0.7518 - val_loss: 0.5220 - val_acc: 0.7513
Epoch 42/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5154 - acc: 0.7507 - val_loss: 0.5218 - val_acc: 0.7492
Epoch 43/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5195 - acc: 0.7483 - val_loss: 0.5212 - val_acc: 0.7476
Epoch 44/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5140 - acc: 0.7519 - val_loss: 0.5210 - val_acc: 0.7519
Epoch 45/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5137 - acc: 0.7521 - val_loss: 0.5206 - val_acc: 0.7504
Epoch 46/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5133 - acc: 0.7524 - val_loss: 0.5250 - val_acc: 0.7488
Epoch 47/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5138 - acc: 0.7519 - val_loss: 0.5187 - val_acc: 0.7510
Epoch 48/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5134 - acc: 0.7523 - val_loss: 0.5319 - val_acc: 0.7428
Epoch 49/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5132 - acc: 0.7525 - val_loss: 0.5208 - val_acc: 0.7507
Epoch 50/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5155 - acc: 0.7523 - val_loss: 0.5569 - val_acc: 0.7311
Epoch 51/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5120 - acc: 0.7532 - val_loss: 0.5189 - val_acc: 0.7510
Epoch 52/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5120 - acc: 0.7526 - val_loss: 0.5177 - val_acc: 0.7522
Epoch 53/100
390870/390870 [==============================] - 29s 75us/step - loss: 0.5143 - acc: 0.7511 - val_loss: 0.5204 - val_acc: 0.7510
Epoch 54/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5175 - acc: 0.7495 - val_loss: 0.5230 - val_acc: 0.7493
Epoch 55/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5123 - acc: 0.7526 - val_loss: 0.5230 - val_acc: 0.7488
Epoch 56/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5128 - acc: 0.7524 - val_loss: 0.5350 - val_acc: 0.7442
Epoch 57/100
390870/390870 [==============================] - 28s 72us/step - loss: 0.5118 - acc: 0.7529 - val_loss: 0.5317 - val_acc: 0.7469
Epoch 58/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5110 - acc: 0.7542 - val_loss: 0.5214 - val_acc: 0.7471
Epoch 59/100
390870/390870 [==============================] - 28s 71us/step - loss: 0.5110 - acc: 0.7539 - val_loss: 0.5240 - val_acc: 0.7492
Epoch 60/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5103 - acc: 0.7543 - val_loss: 0.5316 - val_acc: 0.7407
Epoch 61/100
390870/390870 [==============================] - 27s 69us/step - loss: 0.5104 - acc: 0.7538 - val_loss: 0.5214 - val_acc: 0.7499
Epoch 62/100
390870/390870 [==============================] - 27s 68us/step - loss: 0.5103 - acc: 0.7541 - val_loss: 0.5351 - val_acc: 0.7428
Epoch 63/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5100 - acc: 0.7542 - val_loss: 0.5160 - val_acc: 0.7526
Epoch 64/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5102 - acc: 0.7543 - val_loss: 0.5191 - val_acc: 0.7527
Epoch 65/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5098 - acc: 0.7543 - val_loss: 0.5195 - val_acc: 0.7500
Epoch 66/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5098 - acc: 0.7541 - val_loss: 0.5231 - val_acc: 0.7504
Epoch 67/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5095 - acc: 0.7546 - val_loss: 0.5178 - val_acc: 0.7528
Epoch 68/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5092 - acc: 0.7546 - val_loss: 0.5310 - val_acc: 0.7401
Epoch 69/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5087 - acc: 0.7551 - val_loss: 0.5192 - val_acc: 0.7500
Epoch 70/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5090 - acc: 0.7546 - val_loss: 0.5217 - val_acc: 0.7496
Epoch 71/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5088 - acc: 0.7550 - val_loss: 0.5210 - val_acc: 0.7502
Epoch 72/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5087 - acc: 0.7549 - val_loss: 0.5282 - val_acc: 0.7473
Epoch 73/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5090 - acc: 0.7548 - val_loss: 0.5350 - val_acc: 0.7392
Epoch 74/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5076 - acc: 0.7550 - val_loss: 0.5198 - val_acc: 0.7528
Epoch 75/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5092 - acc: 0.7553 - val_loss: 0.5231 - val_acc: 0.7496
Epoch 76/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5074 - acc: 0.7556 - val_loss: 0.5198 - val_acc: 0.7512
Epoch 77/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5079 - acc: 0.7553 - val_loss: 0.5212 - val_acc: 0.7504
Epoch 78/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5110 - acc: 0.7550 - val_loss: 0.5387 - val_acc: 0.7505
Epoch 79/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5082 - acc: 0.7559 - val_loss: 0.5262 - val_acc: 0.7460
Epoch 80/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5081 - acc: 0.7551 - val_loss: 0.5198 - val_acc: 0.7502
Epoch 81/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5176 - acc: 0.7486 - val_loss: 0.5226 - val_acc: 0.7471
Epoch 82/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5117 - acc: 0.7525 - val_loss: 0.5283 - val_acc: 0.7431
Epoch 83/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5130 - acc: 0.7523 - val_loss: 0.5443 - val_acc: 0.7500
Epoch 84/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5188 - acc: 0.7455 - val_loss: 0.5171 - val_acc: 0.7527
Epoch 85/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5097 - acc: 0.7541 - val_loss: 0.5418 - val_acc: 0.7437
Epoch 86/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5095 - acc: 0.7544 - val_loss: 0.5177 - val_acc: 0.7529
Epoch 87/100
390870/390870 [==============================] - 27s 70us/step - loss: 0.5082 - acc: 0.7550 - val_loss: 0.5190 - val_acc: 0.7515
Epoch 88/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5073 - acc: 0.7560 - val_loss: 0.5881 - val_acc: 0.7489
Epoch 89/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5073 - acc: 0.7563 - val_loss: 0.5170 - val_acc: 0.7505
Epoch 90/100
390870/390870 [==============================] - 26s 68us/step - loss: 0.5073 - acc: 0.7556 - val_loss: 0.5199 - val_acc: 0.7511
Epoch 91/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5070 - acc: 0.7558 - val_loss: 0.5236 - val_acc: 0.7519
Epoch 92/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5073 - acc: 0.7557 - val_loss: 0.5192 - val_acc: 0.7510
Epoch 93/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5066 - acc: 0.7564 - val_loss: 0.5225 - val_acc: 0.7478
Epoch 94/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5062 - acc: 0.7567 - val_loss: 0.5167 - val_acc: 0.7525
Epoch 95/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5058 - acc: 0.7565 - val_loss: 0.5262 - val_acc: 0.7433
Epoch 96/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5059 - acc: 0.7566 - val_loss: 0.5157 - val_acc: 0.7535
Epoch 97/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5055 - acc: 0.7569 - val_loss: 0.5194 - val_acc: 0.7516
Epoch 98/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5059 - acc: 0.7561 - val_loss: 0.5254 - val_acc: 0.7454
Epoch 99/100
390870/390870 [==============================] - 26s 66us/step - loss: 0.5057 - acc: 0.7567 - val_loss: 0.5354 - val_acc: 0.7492
Epoch 100/100
390870/390870 [==============================] - 26s 67us/step - loss: 0.5054 - acc: 0.7567 - val_loss: 0.5189 - val_acc: 0.7509

<keras.callbacks.History at 0x1b77226d3c8>

# Save trained model
classifier.save("../models/trained_deep_neural_network_real.h5")

# evaluation of the model training
evaluation = classifier.evaluate(train_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']), train_set['likes'])
print(evaluation)

558386/558386 [==============================] - 17s 30us/step
[0.5069156529383092, 0.7561507630928384]

del train_set
test_set = _pd.read_pickle('../dataset/m2_n9/model_test_set_3.pickle')
test_set = test_set.drop(columns=[*_cols_bin, *_cols_std])
test_set.shape

(153993, 754)

prediction = classifier.predict(test_set.drop(columns = ['likes', 'stars_review', 'review_id', 'user_id', 'business_id']))
binary_prediction = _binarize(prediction, threshold = 0.5)
binary_prediction

array([[1.],
       [0.],
       [1.],
       ...,
       [0.],
       [1.],
       [0.]], dtype=float32)

set(binary_prediction[:,0])

{0.0, 1.0}

print("Report for Deep Learning approach:")
print(_classification_report(test_set['likes'], binary_prediction))

print("Accuracy for Deep Learning approach:", _accuracy_score(test_set['likes'], binary_prediction) * 100)

Report for Deep Learning approach:
              precision    recall  f1-score   support

           0       0.70      0.39      0.51     50930
           1       0.75      0.92      0.83    103063

    accuracy                           0.74    153993
   macro avg       0.73      0.66      0.67    153993
weighted avg       0.74      0.74      0.72    153993

Accuracy for Deep Learning approach: 74.43974726123915

matrix = _confusion_matrix(test_set['likes'], binary_prediction)
print(matrix)

[[20079 30851]
 [ 8510 94553]]

# draw ROC curve
fpr, tpr, thresholds = _roc_curve(test_set['likes'], binary_prediction)

_plt.plot(fpr, tpr)
_plt.xlim([0.0, 1.0])
_plt.ylim([0.0, 1.0])

_plt.title("Neural Network ROC curve")
_plt.xlabel("False Positive")
_plt.ylabel("True Positive")

_plt.grid(True)
_plt.show()

_del_all()

	review_id	user_id	business_id	stars	useful	funny	cool	text	date
3	yi0R0Ugj_xUx_Nek0-_Qig	dacAIZ6fTM6mqwW5uxkskg	ikCg8xy5JIg_NGPx-MSIDA	5	0	0	0	Went in for a lunch. Steak sandwich was delici...	2018-01-09 20:56:38
5	fdiNeiN_hoCxCMy2wTRW9g	w31MKYsNFMrjhWxxAb5wIw	eU_713ec6fTGNO4BegRaww	4	0	0	0	I'll be the first to admit that I was not exci...	2013-01-20 13:25:59
6	G7XHMxG0bx9oBJNECG4IFg	jlu4CztcSxrKx56ba1a5AQ	3fw2X5bZYeW9xCz_zGhOHg	3	5	4	5	Tracy dessert had a big name in Hong Kong and ...	2016-05-07 01:21:02
7	8e9HxxLjjqc9ez5ezzN7iQ	d6xvYpyzcfbF_AZ8vMB7QA	zvO-PJCpNk4fgAVUnExYAA	1	3	1	1	This place has gone down hill. Clearly they h...	2010-10-05 19:12:35
10	kbtscdyz6lvrtGjD1quQTg	FIk4lQQu1eTe2EpzQ4xhBA	8mIrX_LrOnAqWsB5JrOojQ	4	0	0	0	Like walking back in time, every Saturday morn...	2011-11-30 02:11:15

	review_id	text	bin_truth_score	real_truth_score
3	yi0R0Ugj_xUx_Nek0-_Qig	Went in for a lunch. Steak sandwich was delici...	1	0.878876
5	fdiNeiN_hoCxCMy2wTRW9g	I'll be the first to admit that I was not exci...	1	0.989000
6	G7XHMxG0bx9oBJNECG4IFg	Tracy dessert had a big name in Hong Kong and ...	1	0.964619
7	8e9HxxLjjqc9ez5ezzN7iQ	This place has gone down hill. Clearly they h...	1	0.867897
10	kbtscdyz6lvrtGjD1quQTg	Like walking back in time, every Saturday morn...	1	0.975404

	review_id	user_id	business_id	stars	useful	funny	cool	date	bin_truth_score	real_truth_score
3	yi0R0Ugj_xUx_Nek0-_Qig	dacAIZ6fTM6mqwW5uxkskg	ikCg8xy5JIg_NGPx-MSIDA	5	0	0	0	2018-01-09 20:56:38	1	0.878876
5	fdiNeiN_hoCxCMy2wTRW9g	w31MKYsNFMrjhWxxAb5wIw	eU_713ec6fTGNO4BegRaww	4	0	0	0	2013-01-20 13:25:59	1	0.989000
6	G7XHMxG0bx9oBJNECG4IFg	jlu4CztcSxrKx56ba1a5AQ	3fw2X5bZYeW9xCz_zGhOHg	3	5	4	5	2016-05-07 01:21:02	1	0.964619
7	8e9HxxLjjqc9ez5ezzN7iQ	d6xvYpyzcfbF_AZ8vMB7QA	zvO-PJCpNk4fgAVUnExYAA	1	3	1	1	2010-10-05 19:12:35	1	0.867897
10	kbtscdyz6lvrtGjD1quQTg	FIk4lQQu1eTe2EpzQ4xhBA	8mIrX_LrOnAqWsB5JrOojQ	4	0	0	0	2011-11-30 02:11:15	1	0.975404

	name	address	cuisine	postal_code	latitude	longitude	review_count	stars	OutdoorSeating	BusinessAcceptsCreditCards	...	Thursday_Close	Friday_Close	Saturday_Close	Sunday_Close	average_stars	num_reviews	average_stars_bin	num_reviews_bin	average_stars_real	num_reviews_real
business_id
QXAEGFB4oINsVuTFxEYKFQ	Emerald Chinese Restaurant	30 Eglinton Avenue W	Chinese	L5R 3E7	43.605499	-79.652289	128	2.5	False	NaN	...	00:00:00	01:00:00	01:00:00	00:00:00	3.703313	0	3.703313	0	3.703313	0
gnKjwL_1w79qoiV3IC_xQQ	Musashi Japanese Restaurant	10110 Johnston Rd, Ste 15	Japanese	28210	35.092564	-80.859132	170	4.0	False	True	...	21:30:00	22:00:00	22:00:00	21:00:00	3.703313	0	3.703313	0	3.703313	0
1Dfx3zM-rW4n-31KeC8sJg	Taco Bell	2450 E Indian School Rd	Mexican	85016	33.495194	-112.028588	18	3.0	False	True	...	01:00:00	01:00:00	01:00:00	00:00:00	3.703313	0	3.703313	0	3.703313	0
fweCYi8FmbJXHCqLnwuk8w	Marco's Pizza	5981 Andrews Rd	Italian	44060	41.708520	-81.359556	16	4.0	False	True	...	00:00:00	01:00:00	01:00:00	00:00:00	3.703313	0	3.703313	0	3.703313	0
PZ-LZzSlhSe9utkQYU8pFg	Carluccio's Tivoli Gardens	1775 E Tropicana Ave, Ste 29	Italian	89119	36.100016	-115.128529	40	4.0	False	True	...	NaT	NaT	NaT	NaT	3.703313	0	3.703313	0	3.703313	0

	num	stars	num_bin	stars_bin	num_real	stars_real
business_id
--1UhMGODdWsrMastO9DZw	22.0	4.227273	14.0	4.214286	16.978214	4.267477
--6MefnULPED_I942VcFNA	38.0	3.157895	32.0	3.218750	33.948759	3.218815
--9e1ONYQuAa-CB_Rrw7Tw	1504.0	4.107048	1216.0	4.088816	1205.738732	4.092415
--DaPTJW3-tB1vP-PfdTEg	40.0	3.650000	33.0	3.575758	34.780500	3.642790
--FBCX-N37CMYDfs790Bnw	122.0	3.737705	90.0	3.611111	93.369750	3.646290

	user_name	average_stars	yelping_since	review	years_of_elite	fans	useful	cool	funny	friends	num_reviews	average_stars_bin	num_reviews_bin	average_stars_real	num_reviews_real
user_id
l6BmjZMeQD3rDxWUbiAiow	Rashmi	3.703313	2013-10-08	95	3	5	84	25	17	2374	0	3.703313	0	3.703313	0
4XChL029mKr5hydo79Ljxg	Jenna	3.703313	2013-02-21	33	0	4	48	16	22	27646	0	3.703313	0	3.703313	0
bc8C_eETBWL0olvFSJJd0w	David	3.703313	2013-10-04	16	0	0	28	10	8	358	0	3.703313	0	3.703313	0
dD0gZpBctWGdWo9WlGuhlA	Angela	3.703313	2014-05-22	17	0	5	30	14	4	12598	0	3.703313	0	3.703313	0
MM4RJAeH6yuaN8oZDSt0RA	Nancy	3.703313	2013-10-23	361	4	39	1114	665	279	5542	0	3.703313	0	3.703313	0

	Chinese_av	Japanese_av	Mexican_av	Italian_av	Others_av	American_av	Korean_av	Mediterranean_av	Thai_av	Asian Fusion_av	...	Chinese_av_real	Japanese_av_real	Mexican_av_real	Italian_av_real	Others_av_real	American_av_real	Korean_av_real	Mediterranean_av_real	Thai_av_real	Asian Fusion_av_real
user_id
---1lKK3aKOuomHnwAkAow	NaN	2.333333	3.666667	3.166667	4.352941	3.9375	4.0	NaN	3.25	3.5	...	NaN	2.035809	3.720609	3.226154	4.524134	3.963953	4.0	NaN	3.134858	3.443446
---PLwSf5gKdIoVnyRHgBA	NaN	NaN	NaN	NaN	3.000000	NaN	NaN	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	3.000000	NaN	NaN	NaN	NaN	NaN
---cu1hq55BP9DWVXXKHZg	NaN	NaN	4.000000	NaN	NaN	NaN	NaN	NaN	NaN	1.0	...	NaN	NaN	4.000000	NaN	NaN	NaN	NaN	NaN	NaN	1.000000
---udAKDsn0yQXmzbWQNSw	NaN	NaN	NaN	NaN	5.000000	NaN	NaN	NaN	4.00	NaN	...	NaN	NaN	NaN	NaN	5.000000	NaN	NaN	NaN	4.000000	NaN
--0RtXvcOIE4XbErYca6Rw	NaN	NaN	NaN	NaN	4.000000	NaN	NaN	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	4.000000	NaN	NaN	NaN	NaN	NaN

	review_id	user_id	business_id	stars	useful	cool	date	bin_truth_score	real_truth_score
0	---j05qHS2X7FkXjjMKKtA	E6Aoz-3s4avfweIjziHjbA	cTbFJzHQzFSX-z3JF4abKQ	5	1	1	2018-11-02 15:01:11	-1	0.229833
1	--4GjusuUCMh24c_oh_cEg	YOOsYiXGEtGFX_wSeZNcww	JytR7WvKyytDQNwOHUzSEg	4	1	0	2018-10-13 00:07:17	1	0.987487
2	--4RpVT5wHJ9AfnZkIC3tw	Bdw4E8jFVd6-CbhrNAJ_EA	-CfFjcCcGGDM9MVH_d42RQ	5	0	0	2018-10-14 04:41:42	-1	0.524751
3	--4vJzoC0m5h-yodXv-qCw	jm4a1GghQ4zLCN3lQGMQUQ	XMPBg6r_LqZhy9Cf-4ZJrA	2	0	0	2018-11-01 19:11:50	1	0.979875
4	--B9JxEb5gY5gAgD2BRhDQ	x_6VdQU3CIdakwHod-dNzA	Fn_IxcCtZl1EoS81sq_s9w	3	1	0	2018-10-31 02:01:04	1	0.662471

	review_id	cuisine_av_hist	cuisine_av_hist_bin	cuisine_av_hist_real
0	---j05qHS2X7FkXjjMKKtA	4.000000	4.285714	3.969038
1	--4GjusuUCMh24c_oh_cEg	3.604167	3.607143	3.610109
2	--4RpVT5wHJ9AfnZkIC3tw	5.000000	NaN	5.000000
3	--4vJzoC0m5h-yodXv-qCw	NaN	NaN	NaN
4	--B9JxEb5gY5gAgD2BRhDQ	NaN	NaN	NaN

	review_id	user_id	business_id	stars	useful	funny	cool	date	bin_truth_score	real_truth_score
0	---HLAnHbuLi7vd5TL6uYg	zyp8SaRnZ94sWZpLrifS1Q	l6xZVTEtdZAvNpL1JhYGuw	4	0	0	0	2018-05-26 13:49:33	-1	0.595406
1	---L4b6VR6HoB-q7cfMWIA	697iJkhX1mkVF9RNhn114Q	XiXu6WHbDoopKpeg7DfKdQ	5	3	1	2	2018-04-15 18:54:27	1	0.993384
2	---sPYSgArT4Sd5v1nDVMQ	iVSuN8PrtKVtLzhNiu23uA	OumGHdbdp7WgyYMhcAdjhw	1	0	0	0	2018-07-07 15:16:09	1	0.519254
3	--0SzSMXVUoAXfackNoB4g	v9P7J6hWWtIblnylQ5UBfA	iCQpiavjjPzJ5_3gPD5Ebg	5	0	0	0	2018-05-09 13:59:37	1	0.983368
4	--1JMhPk6K9fZo4FOp_yMw	2xZ1mHP14as5RJ1KOrVU4A	QJatAcxYgK1Zp9BRZMAx7g	2	0	0	0	2018-06-16 17:00:42	1	0.866956

	review_id	cuisine_av_hist	cuisine_av_hist_bin	cuisine_av_hist_real
0	---HLAnHbuLi7vd5TL6uYg	NaN	NaN	NaN
1	---L4b6VR6HoB-q7cfMWIA	3.013889	3.014925	3.039444
2	---sPYSgArT4Sd5v1nDVMQ	NaN	NaN	NaN
3	--0SzSMXVUoAXfackNoB4g	1.000000	1.000000	1.000000
4	--1JMhPk6K9fZo4FOp_yMw	2.750000	2.666667	2.748879

	review_id	user_id	business_id	stars	useful	date	bin_truth_score	real_truth_score	cuisine_av_hist	cuisine_av_hist_bin	cuisine_av_hist_real	coll_score	coll_score_bin	coll_score_real
0	hhVr1uH7XaRlbIHI8dYvbA	FYhU1fKQ7n11WQ7gcFYOag	--1UhMGODdWsrMastO9DZw	5	0	2018-01-11 19:55:31	1	0.622302	NaN	NaN	NaN	NaN	NaN	NaN
1	no4Eo4WloZRLwcYZP9gfhg	m-p-7WuB85UjsLDaxJXCXA	--1UhMGODdWsrMastO9DZw	5	0	2018-02-25 17:47:12	1	0.964784	4.000000	4.000000	4.000000	3.757630	3.808050	3.714375
2	8QUwzeXeyJ3L15lKfhKLsQ	Bsy9F-59sl9OT_bvZNl3hA	--1UhMGODdWsrMastO9DZw	1	0	2018-05-06 04:22:48	1	0.871544	3.000000	3.000000	2.990709	2.926486	2.974252	2.874015
3	BIecLw546kAlD7kmlk7vXA	sTVGcezME7gYBhIlYtcfpg	--1UhMGODdWsrMastO9DZw	2	0	2018-04-22 17:42:09	1	0.988395	NaN	NaN	NaN	NaN	NaN	NaN
4	0nVZ9Cg1A1lVA8EFWbt5lg	EisUuXVeVJN_FcFiE-tqwA	--6MefnULPED_I942VcFNA	3	2	2018-05-21 05:09:07	1	0.927789	2.966667	2.923077	2.954478	2.985799	2.973742	2.915633

	review_id	user_id	business_id	stars	date	bin_truth_score	real_truth_score	cuisine_av_hist	cuisine_av_hist_bin	cuisine_av_hist_real	coll_score	coll_score_bin	coll_score_real
0	-j8YU0f5cL_fbnzsi1zkpA	rEsBrt6U7i8O4rC81lV6NQ	--6MefnULPED_I942VcFNA	4	2018-10-21 18:45:39	1	0.997555	NaN	NaN	NaN	NaN	NaN	NaN
1	1vEQsKjTfGCcNdQ2Hhuhew	yEP9vNFq3edLldNzhm6hgQ	--6MefnULPED_I942VcFNA	5	2018-09-21 01:07:38	-1	0.553523	1.8	2.000000	1.799679	1.838975	2.007105	1.777964
2	P6ZgOVwyGlvis4m2Cra13g	uH8tTLb3Fz64GtEMWpZZcQ	--6MefnULPED_I942VcFNA	5	2018-09-09 03:20:03	1	0.990602	4.3	4.333333	4.299574	4.349620	4.302949	4.288981
3	ap-_EXUS49YhyukC23p2Gw	NQhvEYuYOa5psBxEoNvq2g	--6MefnULPED_I942VcFNA	1	2018-10-04 01:37:05	1	0.968214	NaN	NaN	NaN	NaN	NaN	NaN
4	tKFDdiQ7rGMsdxgvIz2Sxg	zbrH2lnZgWicvwoXR0qcSg	--9e1ONYQuAa-CB_Rrw7Tw	5	2018-11-06 19:48:01	1	0.995667	NaN	NaN	NaN	NaN	NaN	NaN

	user_id	business_id	tips_date
0	Fzz-0v1yHLaWuTV64b1miA	EZZjaiV8ik05NUepqdeP2A	2018-02-14
1	AuSB69SSnaPNTwwbFk93MQ	bAHDSbpJE3kKJkW9OBgOyw	2018-03-23
2	ouk36OGbx25nO23b10L5jw	THO77IL6DLob9Agt9QCjsw	2018-03-22
3	gwmyGLz4eBm9QiBU_Ze2KQ	RJOFGZZf3ho04ku0fcFRdA	2018-03-29
4	BzcdTNAe_jtXfnXFdFYQsA	_pBXtjN43eqMV0XZTz7nmw	2018-04-02

	user_id	business_id	tips_date
0	Yr0B0aVb94i2oIm1gLXgfg	yGDiAVoQB8LX3OsJ4e2I0A	2018-10-05
1	z6qIyc-_oIbgDWAembZ04w	wFO5HLn-GSfYCSPPcbyqoA	2018-10-12
2	EyYVD9n7PlYYLTSEQ5t14w	yhDAzBBjFujZbHwBPfE2eQ	2018-10-15
3	90fhUaWIY6ctVAX5jQ2GNQ	ZMJOURno1xJS7PG3ZhfeyQ	2018-10-20
4	k7J0CjxFoxdSayhlDE-k7w	X3qrbOkhdCjm0NTBX7T80Q	2018-10-30